Preliminary Version, comments are welcome

Real Wages and Skill Premia in China, 1858-1936* [Job Market Paper] Se Yan UCLA [email protected] Abstract: Under pressure from foreign governments in the late nineteenth and early twentieth centuries, China was transformed from a closed agrarian economy to an open economy undergoing industrialization. To date, scholars have had relatively little data to measure the impact of these changes on China’s population. This project estimates nominal wages for workers in nearly fifty Chinese cities from 1858 to 1936 using the employment records of the China Maritime Customs service. It also constructs cost of living indices on the basis of price data and expenditure weights developed from the customs service’s trade statistics and budget surveys. The resulting nominal wage series and cost of living indices make it possible to estimate real income trends for different groups of workers. This paper finds that skill premia rose rapidly during the first two decades of industrialization. After the 1910s, the wage gap between skilled and unskilled labor began to decline, while the gap between highly skilled and unskilled labor leveled off. These changes in the skill premia were driven by movements in the wages of skilled and highly skilled labor. China’s enormous reservoir of unskilled labor kept unskilled wages stagnant until the last ten years of the period, when the real wages of all three groups began to grow rapidly. This paper explores possible explanations for the patterns in the skill premia, including changes in the supply and demand of skilled and unskilled labor brought by technological and educational progress, as well as external trade shocks caused by events such as World War I.

* I devote my greatest thanks to Ken Sokoloff for his invaluable support and guidance. I am also indebted to Naomi R. Lamoreaux, Dora L. Costa, Jean-Laurent Rosenthal, Matthias Doepke, Peter H. Lindert, R. Bin Wong, Hongbin Cai, Debin Ma, Ramon H. Myers, Hans van de Ven, and participants of UCLA economic history seminar, the 2007 Cliometric Conference, and the 2006 and 2007 All-UC economic history conferences. I acknowledge financial supports from Economic History Association graduate fellowship, UCLA Chinese studies fellowship, All-UC economic history fellowship, UCLA Economic History Center fellowship, and research grants from Ken Sokoloff and Ramon H. Myers. Archival assistance from the Second Historical Archive of China is also appreciated. All errors are mine.

1. Introduction In recent years there has been great interest among economists in the record and processes of economic development of China in the late nineteenth and early twentieth centuries, when China first opened its economy and began to industrialize. Prior to the 1840s China was a closed, largely agrarian economy. Under pressure from various foreign nations led by Great Britain, China gradually opened its economy to foreign trade and, much later on, foreign direct investment. By the early twentieth century, economic openness ushered in a rapid industrialization and commercialization in coastal China. Expanding industrial sectors rapidly substituted imports, and even began to export. This snap-shot picture of China’s accelerating economic change has been widely shared and understood by historians and economists. What is less understood, however, is how these economic changes affected people’s real income and standard of living. This paper studies trends in real wages and wage inequality in China in the late nineteenth and early twentieth centuries. I construct nominal wage series from the records of employees in the China Maritime Customs service (hereafter “CMC”) for nearly fifty Chinese cities from 1858 to 1936. I also construct cost of living indices from price data and household budget information contained in CMC trade statistics, surveys and other sources of surveys. The resulting nominal wage series and cost of living indices make it possible to estimate real wage trends and skill premia for different groups of Chinese workers. To my knowledge, this is the first time that such real wage series in China from 1858 to 1936 is produced, and is also the first time that this unique archival source is systematically tapped. Using these newly constructed real wage series, I find that skill premia rose rapidly during the first two decades of industrialization. After the 1910s, the wage gap between skilled and unskilled labor began to decline, while the gap between highly skilled and

1

unskilled labor leveled off. These changes in the skill premia, I show, were driven by movements in the wages of skilled and highly skilled labor. China’s enormous reservoir of unskilled labor kept unskilled wages stagnant until a rapid growth in the last ten years of the period. By contrast, supplies of skilled and highly skilled labor were extremely scarce, driving up their real wages and the skill premia in the first few decades of industrialization. After the 1910s, however, progress in education increased the supply of skilled labor. Skilled wages started to decline, but wages for highly skilled labor stayed relatively high because the supply of this type of labor remained insufficient. The decline of skilled wages after the 1910s was also possibly resulted from external shocks such as the WWI and a series of civil wars. My findings in this paper contribute to an ongoing effort to understand the dimensions of China’s industrialization in the early twentieth century. Scholars have been trying to put together the output data of the entire nation as well as important sectors and regions, and many evidences presented in these studies show remarkable achievements in China’s industrialization in this period (Chung-li Chang 1962; Liu and Yeh 1965; Chang 1969; Rawski 1989; Ma 2006). Despite these strenuous efforts trying to fill the data gap, a consistent long-run series of macroeconomic indicators such as output, real income, wages and prices, could never be constructed due to lack of national income accounting. My study goes beyond these efforts to measure the effects of these economic changes upon real wages and standard of living and suggests that the rapid economic changes started to bring a rapid and sustained growth to real wages as late as from the mid 1920s. This can contribute to scholars’ efforts to map the contour of China’s industrial development. Similar to the work of Lindert and Williamson (1983) and Margo (2000), this paper studies wage inequalities with a unique wage database covering about fifty Chinese cities from 1858 to 1936. The findings of this paper can also help us understand long-run

2

inequality pattern. A large body of academic work has shown both theoretically and empirically that technological changes in the first and the second industrial revolutions and in more recent periods had different skill biases and the affected inequality patterns differently (Griliches 1969; James and Skinner 1985; Mokyl 1990; Katz and Murphy 1992; Goldin and Katz 1996; Goldin and Katz 1998; Caselli 1999; Krusell et all 2000; Acemoglu 1998, 2002a, 2002b, 2003; O'Rourke, Rahman and Taylor 2007). Scholars have also observed that the expansion of mass education contributed to a decline in income inequality in the US in the twentieth century (Goldin and Margo 1992; Katz and Murphy 1992; Autor, Katz and Krueger 1998; Goldin and Katz 2001, 2001; Card and Lemieux, 2001; Acemoglu 2002). External shocks, such as wars, regime changes, minimum wage regulations, etc. are also found to greatly affect wage and income inequalities in the US and France (Card and DiNardo 2002; Piketty, Postel-Vinay and Rosenthal 2004; Piketty and Saez 2001, 2006; Lemieux 2006). However, the empirical evidence bolstering these arguments is largely from leading countries such as Britain, America and France, and the study of inequality patterns in developing countries in early periods remains weak. Follower countries often have a much larger stock of unskilled labor and a smaller population of skilled labor, and this difference in factor endowments between leaders and followers may lead to completely different trends in wage inequality during periods of technological and educational advance. The newly constructed real wage series and skill premia in China from 1858 to 1936 and the explorations of the driving forces shaping the wage inequality patterns can contribute to a deeper understanding of existing literature. The rest of this paper is organized as follows. Section 2 plots a brief picture of the institutional and economic changes in modern China. Section 3 introduces the primary data source, the CMC archives and studies the representativeness of the dataset. In Section 4 I

3

present the nominal wage series obtained from the hedonic regression on the sample dataset I have. In Section 5 the cost of living indices for different income groups are presented. In Section 6 the real wage estimation is presented based on nominal wage series and cost of living indices. Section 7 studies the changes in skill premia and its causes, followed by a conclusion in Section 8.

2. Background This paper studies trends in real wages and wage inequality in China in the late nineteenth and early twentieth centuries, the period when China first opened its economy and began to industrialize. Prior to the 1840s China was a closed, largely agrarian economy. Under pressure from various foreign nations led by Great Britain, China gradually opened its economy to international trade. The Treaty of Nanking in 1842 granted foreigners access to five open ports and stipulated a nominal 5 percent ad valorem tariff on all goods leaving and entering China. 1 By the early twentieth century, the number of cities open to trade had climbed to 48 cities, and the volume of trade began to increase at a rapid pace (see Figure 1). From 1871-84 to 1920-29 exports as percentage of GDP increased from 2.5 percent to 7.3 percent, the per capita value of foreign trade grew from 0.58 to 3.01 U.S. dollars, and China’s share of world trade increased from 1.3 percent to 2.4 percent.2 China’s defeat in the Sino-Japanese war in 1895 ushered in a second wave of economic change. The Treaty of Shimonoseki forced China to allow Japanese businesses to invest directly in China and produce goods and services that could be sold to other nations as well

Except for a few special commodities such as opium and tea, the tariff rates for both imports and exports were fixed at five percent ad valorem until China 1929. However, because commodity prices rose rapidly after 1858, the effective tariff rates were often below 3 percent and were never above 4 percent. See Cheng (1956) p. 10 and p. 53. 2 See Dernberger (1975). p. 27. 1

4

as marketed within China. This privilege was quickly extended to other foreign powers by dint of the most-favored-nation clauses included in their trading agreements with China. Foreign capital poured in to finance enterprises in the railways, telecommunications, shipping, and manufacturing sectors.3 It was during this second wave of economic change that the industrialization of China really began. Cotton textiles were one of the most rapidly growing industries. Trade records show that domestic yarn firms started to export in 1913 and by the mid 1920s had largely displaced imports (see Figure 2). Although the scarcity of data made it virtually impossible to construct annual time series of GDP and other major economic indicators, 4 scholars have made various estimations on the speed and magnitude of industrial expansion and economic growth, most of which espouse this snap-shot picture. For example, John Chang (1969) constructed an industrial output index which indicates that industrial output grew 9.8 percent annually from 1912 to 1936. Liu and Yeh (1965) estimate an annual growth of per capita GDP at 0.33 percent, while Rawski (1989) contends that per capita GDP growth in advanced regions of China had already attained a similar rate to that of Japan in the Republican era from 1912 to 1936 (for a detailed summary see Ma 2006). The benefits of industrial expansion and economic growth in the early twentieth century should have trickled down to people’s income growth. But without sound empirical evidences, scholars have very bifurcated views on whether there was really sustained income growth and how different income groups of people benefited differently. This situation calls for the collection of new information, which would make it possible to construct new wage

For information on foreign investments in China, see Hou (1965), chapter 3 and Dernberger (1975) p. 28. The historical record for Chinese economiy mainly comes from the records kept by China’s rulers, officials, and local elites. These records, such as tax records and land contracts, however, reveal more about how and why the Chinese people were organizing their lives to establish a special kind of civilization but very little quantitative information about what China’s economic performance was all about. 3 4

5

series and cost of living indices, and contribute to a better measurement of the speed and magnitude of economic changes in China in the late nineteenth and early twentieth centuries.

3. Data 3.1

China Maritime Customs and Its Archives and Publications

My wage and price data are collected from CMC archives and publications. These archives are among the richest resources for the study of trade, economics, politics, society, international relations, and many other aspects of Chinese history from the 1850s to 1948. The CMC was answerable to the Chinese central government, but its top administration consisted largely of British and later Japanese and American officials and a major proportion of its middle-rank technical and management staff were foreigners. The CMC was probably the only bureaucratic organization in China that continued to operate uninterruptedly throughout the period from 1859 to 1948. The CMC’s primary tasks, of course, were collecting revenues from, and recording and publishing data on, China’s foreign trade. But it took on more and more functions over time, including r collecting revenues for domestic trade, administering the postal system, developing inland and coastal waterways, and representing China at international fairs. The CMC’s geographical reach grew from just fourteen stations during the 1860s to nearly fifty during the 1920s, covering not only the coastal regions but also inland cities including those near the border with Burma and along the Amur River on the northern tip.5

5 For a detailed introduction, see Hans van de Ven (2002), Notes from the field: The Maritime Customs Service Project and the website of Chinese Maritime Customs Service Project, Department of Historical Studies, University of Bristol: http://www.bris.ac.uk/history/customs/

6

The most well-known and studied of CMC’s publications are its trade statistics.6 These trade statistics record the quantities and the values of all commodities passing through each treaty port and therefore provide a precious source for the study of trade patterns and commodity prices in these cities. The CMC also published annual and decennial analytical reports on economic activities in those port-cities, as well as in the provinces where they were located. 7 These reports provide extremely rich information on many aspects of the Chinese economy, such as trade and shipping, the natural environment, agriculture, industry, currency and finance, mines and minerals, communications, education, administration, justice and police, health and sanitation, and population, and are of enormous values for scholars of these interests. Compared with its published trade statistics and reports, the CMC’s archives have been much less studied. They consist of about 55,000 volumes stored in the Second Historic Archive of China in Nanjing. Around half these volumes pertain to labor. They include surveys of local wages and standards of living, CMC wage scales, and most importantly, the Service Lists – that is, the individual personnel records of CMC employees. The CMC started to keep the Service Lists in 1875 and continued until the end of the service in 1948, maintaining the series in almost the same format and with almost the same contents for this entire time. In each year the Service Lists recorded each employee’s name, home town, year of joining the service, year of being promoted, year transfer to the current customhouse, rank, and monthly salary.

3.2

My Wage Data

See, for example, Cheng (1956), Hsiao (1974), Lyons (2003) The most well-known and cited reports are the so-called “Decennial Reports”, which are a series of publications covering a wide range of natural and social conditions in all the treaty ports in China from 1882 to 1931. The annual publications, “Returns of Trade”, also contain reports on trade and economy in each treaty port.

6 7

7

The Service Lists are the primary sources for my wage data. However, the CMC did not start compiling the Service Lists until 1875 and stopped reporting wages in the 1930s, so I must supplement the Service Lists with other information from the CMC’s personnel management archives. For the more recent period I use the Service Correspondences and the Memoranda of Services. These records provide information on temporary employees

not

covered by the Service Lists. Unfortunately, these records are not available for the earlier period because the archives in which they were stored were destroyed during the Boxer Rebellion in 1900. The only sources available for the period before 1870 are correspondence and dispatches between the CMC’s head office and local branches. Although data on wages were scattered through these documents, I have been able to garner hundreds of data points from them and to extend my wage series back as far as 1858.8 My dataset stops at 1936 because the Sino-Japanese War broke out in 1937, and the economy was disrupted by war from then until the Communist victory in 1949. The complete dataset contains 44,600 observations.9 Each observation records, for one particular year from 1858 to 1936, the wage received by a Chinese employee of the CMC, as well as biographical information about the employee, such as name, job title, home town, current port, the year beginning the service, year promoted to current position and year transferred to the current port. I excluded foreign workers from the dataset because their wages were significantly higher than those in local Chinese markets. The summary statistics reported in Table 1 show the geographic coverage of the dataset. The CMC treaty “ports” were not just coastal or river cities: 12,700 data points came from 8

There are several hundred volumes of dispatches and correspondences in early periods, many of which contain precious information on the CMC’s internal management and its observations of local society. I have only examined a small part of these documents. Given more time I will be able to collect more wage records and other precious information. 9 The same employee often shows up in the dataset for a few times in different years. This feature of the dataset allows tracking the career paths of the CMC employees in future researches.

8

inland cities, constituting 28.48 percent of the whole sample. Some of these inland ports were along the upper and middle Yangtze River, some in northernmost Manchuria, some in the northwestern Shanxi and Shaanxi provinces, some in the southwest Yunnan province and some near the borders of Burma and Vietnam. There is even one Tibetan treaty port, Yatung. I categorized these cities into nine regions (see Appendix A1). The CMC’s personnel hierarchy was very complex. The original record contains more than 100 specific positions, each further subdivided into several ranks. My wage dataset focuses only on the 35 most common occupations, which I categorize into four groups (unskilled, skilled, highly skilled and senior officials) according to their human capital requirements and the types of services provided (see Appendix 2).10

3.3

Data Coverage and Representativeness

The purpose of this paper is to estimate the long-run trend of wages for different income groups in China, utilizing the wage data developed from the CMC internal personnel archives. Since all these data points are the wages earned by the CMC employees, this methodology has several intrinsic problems. First of all, the CMC employees all lived in cities. So to the extent that labor markets are not in equilibrium, their incomes do not tell us much about trends in wages in rural areas. 11 Furthermore, only treaty ports, not all cities, were included in the dataset. Treaty ports were usually large and middle cities, and were more commercialized and integrated into national

Due to the nature of the CMC system, its personnel distribution was highly skewed toward skilled labor. Rural income and economic development have been extensively studied by quite a few scholars, such as Dwight Perkins, Ramon Myers, Philip Huang, Loren Brandt etc. However, studies on urban income and wages are very scarce. The CMC’s annual and decennial reports also include studies on rural incomes and standard of living. 10 11

9

and even world markets. It is likely that wages in the treaty ports might be significantly different from those in other rural areas. Second, the wages paid by the CMC might have deviated from those in the local labor markets. Certainly, foreign employees received wages much higher than their Chinese colleagues. Their wages were not able to reflect any information of the wages in local markets. In addition, the CMC’s senior officials were usually paid extremely high. Their wages were not able to reflect those paid by private-sector firms. Instead, they were analogous to local government officials in terms of income. 12 Therefore observations of these employees are excluded from my sample. The question here is, whether the wages of other Chinese employees, such as coolies, engineers, mechanics, and clerks, were able to reflect market wages. Because the CMC’s wage system did not adjust instantaneously to reflect labor market fluctuations and rising living cost, it is possible that temporarily the CMC paid less than the market wage. Certainly, the CMC would not have been able to hire if in the long run, it had paid wages that, all things being equal, were below those comparable in the cities in which they operated. Existing archives in the early twenty century show that CMC officials were fully aware of market wages and made concerted efforts to match them. The Inspector General’s Circular No. 3002 in 1920 clearly shows such efforts.13 Because the “pay of the Chinese outdoor staff … has been found insufficient in these days of increased cost of living

12 I did a rough comparison of salaries received by senior the CMC officials and incomes of senior gentryofficials estimated by Chang Chung-li (1962), and found them in the same level. 13 Circulars were issued by the Service’s head, the Inspector General. They were confidential documents to which only senior Customs officials such as Customs Commissioners had access. Like Imperial Edicts, they were law until explicitly superseded by a new Circular. They were the key texts of the Customs Service. Circulars tell us how the Custom Service organized itself, discharged its routines, and responded to events. For details, see Bickers and van de Ven (2007), China and the West: The Maritime Customs Service Archive from the Second Historical Archives of China, Nanjing, (London: Thomson Gale)

10

to attract good men or to retain desirable employees,”14 the CMC’s head office instructed its branches to survey the market wages and living costs very carefully: “as local conditions vary so much at different ports, what would be adequate at one being inadequate or excessive at another ... I have to request you therefore to forward: a return showing the local average wages and average expenses of men performing work similar to that performed by the corresponding Customs employee”15 Another important document, Circular No. 3839 in 1928, expresses more explicitly the CMC’s efforts to match its wage to the market wage and accommodate the rising cost of living: “In order that I may be in a position to understand clearly whether the pay at each individual port of the lower grade ranks of employees is on a scale which takes into account the actual cost of living as well as the current market value of the services rendered by such employees, I have to instruct you to prepare and forward to me … a statement … showing the average market prices prevailing at your port during the previous six months for the various necessaries of life, as well as the average wage paid by Chinese and /or foreign firms at your port to the specified classes of employees. ”16 On the other hand, the question remains that the CMC, as a special bureaucratic organization, might have paid significantly higher than the market wage in order to attract best people. This question can be answered by carefully comparing the wages from the CMC and from the local private-sector firms. Fortunately, the CMC conducted six nationwide wage surveys in the three years from 1929 to 1931. These surveys compared wages of employees from the CMC’s local customhouses, local Chinese firms, and local foreign firms. Table 2 reports the mean wages of observations from the 1929-31 surveys by dates, locations 14 and 14

16

the Inspectorate General of Customs, Beijing, 1920, Circular No. 3002, CMC Archive No. 679(1)-16231

the Inspectorate General of Customs, Beijing, 1928, Circular No. 3839, the CMC Archive No. 679(1)-16228

11

and occupations of these observations. This table revealed mixed information: the wages from three types of employers do not seem to differ too much with each other by dates and locations, but the wage discrepancies by occupations were conspicuous. Therefore a better study is needed to show whether the CMC had successfully paid their employees fair marketlevel wages. To do so, I run the following regressions for the survey data 3

3

9

log(wage) = α + ∑∑ βi , j ⋅ typei ⋅ skill j + ∑ γ k ⋅ regionk + ε i =1 j =1

(1)

k =1

The variable “type” indicates different types of wages. Three types of wages were collected in the 1929-31 surveys: wages of Chinese firms, wages of local foreign firms, and the CMC wages. The variable “region” indicates the regions these wage observations came from. The 1929-31 surveys contain 25 occupations. These occupations are categorized into three groups: unskilled, skilled and highly skilled, which are indicated by the variable “skill”. The categorization follows the criteria used for my main wage data, and the detailed criteria can be found in Appendix A2. This specification includes an interaction term of types and skills. It allows me to compare the differentials of the CMC wages and those of privatesector firms by three skill groups. Tables 3 reports the regression coefficients, which indicate the log wage differences compared with the base group, which is unskilled CMC wage. It is clear that that, for the unskilled and highly skilled labor, wages from the CMC and foreign firms were statistically indistinguishable, while for skilled labor foreign firms paid about 6.6 percent higher. This was not a big gap, and it is possibly because the occupations of skilled labor from these employers are highly heterogeneous. Therefore, the regression strongly suggests that the wages from the CMC and local foreign firms do not differ from each other.

12

Table 3 also indicates that the CMC and local foreign firms paid significantly higher than local Chinese firms. The differences between the CMC wages and the wages of local Chinese firms were 22.2, 11.3 and 7.6 percent for three skill groups respectively. The wage differences between local Chinese and foreign firms were also significant. However, this result is somewhat illusory. Chinese firms generally provided food, accommodations, and other non-pecuniary benefits to their employees, while the CMC and foreign firms seldom did so. Such a difference in compensation method has been thoroughly discussed in the reports from local customhouses. 17 According to these survey reports, the board and accommodation benefits usually value 12 to 20 percent of the monetary income.18 With such large in-kind benefits, the total incomes received by employees in Chinese firms were no lower than those worked in foreign firms and the CMC. To conclude the study, because the CMC was a large bureaucratic organization that was not able to adjust wage scales instantaneously, it is possible that temporarily the CMC wages might have been different from the market level. However, the CMC tried hard to set its wage scales to match the market equilibrium level, and the CMC’s survey data also suggest that the wages paid by the CMC, local Chinese firms and foreign firms did not statistically distinguish from each other. This study is in favor of the view that local labor markets were highly integrated. Different types of employers were competing for skilled and unskilled laborers, and equilibrium wages were attained from the labor market competition.

17 For example, a large number of surveys from local customhouses indicate that wages from Chinese firms did not include “free food and housing, value $7” (China Maritime Custom Archive No. 679(1)-16231 and 16232, “Ports Returns of Living Expenses, etc, of Native Outdoor Staff (Excluding Chinese Tidewaiters) , Submitted in Reply to Circular No.3002”); also see, for example, most Chinese firms provided “lodging and boarding”, or “mess allowance valuing $3 to $4”, or sometimes even “food, clothing, quarter, and festival bonus” (China Maritime Custom Archives No. 679(1) - 16233, 16234, 16235, 16236, 16237, 16238, “Cost of Living Returns for Half Year Ending 31th March, 1929 to September 1931”) 18 See the same archives

13

4. Nominal Wages A major goal of this paper is to construct annual time series of nominal and real wages by skill intensities and estimate the trend of skill premia between different skill groups. 19 A main problem in doing so is to adjust for changes in the attributes of the sample over time – for example, the location of the customhouse. The solution that I adopt is the method of hedonic regression. In this hedonic regression, the dependent variable is logarithm of the wage earned by a particular worker, and the independent variables are characteristics of the worker under analysis: the year and the location of this observation, this worker’s job title or job category, and his length of service (the variable “tenure” and its second-order term). The regression reveals the prices of these characteristics by identifying them with the regression coefficients. 20 The regression specification is followed below: 3

21

log(wagei ) = α + ∑∑ δ k ,t ⋅ weightm,t ⋅ job _ categoryk ⋅ yeart k =1 t =1

9

+ ∑ φ j ⋅ region j + γ 1tenurei + γ 2tenurei + ε i 2

(2)

j =1

i = 1,...,44600; m = 1,...,9 where log(wagei ) is the log of the monthly wage pertaining to observation i; α is the constant term; job _ categoryk indicates three job categories: unskilled, semi-skilled, and

This nominal wage dataset can certainly be used in various studies, such as the general trend of wages, the trend of regional wage differences, and I have done a pilot study on the trend of regional wage differentials. In this study I divide regions into two categories: inland or coastal regions. The study shows that regional differentials were small in early periods. The gap expanded very fast since economic openness in the 1860s, and particularly since the 1890s when the ban on foreign direct investment was lifted. The regional gap started narrowing rapidly during the WWI, and leveled off in the 1920s and 1930s. This result shows that economic openness and industrialization impacted coastal regions first and raised the wage levels there. The vast inland regions were impacted by industrialization and globalization much later, and the wage levels in the inland began to catch up since the 1910s. A detailed study of regional income differentials will be another chapter in my dissertation. 20 Robert Margo has a detailed explanation of this method (Margo 2000). 19

14

skilled. 21 yeart indicates 21 three-year periods from 1858 to 1936; region j indicates 9 regions; tenurei indicates years of service this observation pertains to, and its quadratic term is also included; and ε is the error term. My current wage dataset contains 35 specific occupations, and they are grouped into three categories mentioned above. The interaction terms of the variables "job_category" and the year generates annual wage series for all three job categories, all compared to the base group: the unskilled wage in the first period: 1858-1875.22 However, because there are 35 occupations, this three-group categorization is very coarse, and the occupations within each category are very heterogeneous. From year to year, the occupation compositions within one job category change a lot. If within each category the variation between occupations is small, the composition change is not a serious issue. But when the occupations within one category are very heterogeneous, changing composition can lead to significant bias. There are a few methods to control the bias resulted from changing occupation composition in this unbalanced panel. The method I use here is to further divide occupations within each category into subcategories. There are 9 subcategories, and the details can be found in Appendix. Then I assign appropriate weights to these subcategories and run a weighted regression. Ideally the real distributions of these occupations in each year should be adopted. As these distributions are not available, I assign equal weights to these subcategories.23

21

Due to the nature of the CMC system, its personnel distribution was highly skewed toward skilled labor, and in particular, office clerks. In addition, as noted earlier, the CMC's senior officials corresponded to government officials in local governments, and their wages were extraordinarily high, as shown in Table 1. Therefore the current study excludes the sample of senior officials. 22 Because wage observations before 1875 are very scarce, they all group into the first period. Starting from 1876, every three years are grouped into one period. Therefore there are 21 periods. 23 The reason that I assign weights to subcategories instead of each occupation is that these 35 occupations do

15

This weighted regression effectively controls the bias from changing occupation compositions. For example, for unskilled labor, there are three subcategories, and each is assigned for (1/3) weight. If there are n observations in the first subcategory, each observation only accounts for (1/(3n)) weight. With this controlling method, even very skewed distribution within one job category does not lead to the wage estimation biased toward that occupation. An alternative method to control this possible bias is to run a regression with 9 subcategory dummies. 24 This method is straightforward, but it is not applicable in my regression specification, simply because I pool three job categories together in one single regression in order to estimate skill premia between job categories. These subcategory dummies and 3 job category dummies are not independent of each other. In the regression I group the sample in three-year intervals except that I group all the observations before 1875 into one group. Although this method is not possible to construct annual wage series, this method is reasonable because the CMC wages intrinsically reflect long-run equilibrium wages and annual variations are not able to be reflected. An alternative estimation method is to estimate annual variations by assigning each year as a dummy. The advantage of this method is that it constructs annual wage series. But when the sample size of each year is not big enough, the coefficients become volatile. So an acceptable way to accommodate this problem is to smooth the annual wage series by moving average. These two methods do not show significant differences.

not show up every year. Some occupations have fewer frequencies and concentrate only in a few periods. When one occupation does not show up, the weight is not applicable. This difficulty makes this method not able to control the bias from changing occupation composition. 24 Margo (2000) uses a similar method. He runs separate regressions for each of three job categories. For each regression, he controls occupation variations by including occupation dummies in the hedonic regression.

16

The coefficients of the interaction dummies are plotted in Figure 3. The figure shows that, unskilled wages declined slowly in the late nineteenth century until about 1890. From 1890 to the mid 1920s the unskilled wages rose gradually, followed by a rapid growth to 1936; the skilled wages had a mild growth from 1858 to about 1910, followed by a brutal drop around the period of the WWI. But it bounced back and grew rapidly from the mid 1920s to 1936; the highly skilled wages show very similar trend to the skilled, and the only conspicuous difference is that the highly skilled wages did not suffer too much during the WWI. There was only a small drop for the highly skilled wages. Silver price was a critical factor driving up and down the nominal wages. Silver, in the forms of sycees and standardized coins, and copper cash, were two legal tenders in modern China until the currency reform in the 1930s. The gradual decline of the unskilled wages in the late nineteenth century was possibly caused by silver appreciation in that era. The gradual rise of nominal wages for all three groups from the last decade of the nineteenth century corresponds to the inflation caused by silver depreciation. Silver kept depreciating rapidly in the 1920s and the 1930s, until around 1934 when the Silver Purchase Act in the US raised the silver price. The inflation resulted from silver depreciation was echoed by rapid wage growth in the 1920s and 1930s. The WWI was a shock to the wage trends. Skilled group suffered most during the WWI. This is reasonable because they were most sensitive to the changes in international trade. Unskilled labor was less sensitive because demand shocks were relatively small compared with almost infinitive supply. Highly skilled labor was always in high demand, and they were also less impacted. Besides currency and war shocks, fundamental changes in the economy and the labor market might have played more critical roles in shaping the wage trends. Industrial

17

expansion in treaty ports generated larger demand for both unskilled and skilled labor. But these critical roles are difficult to discern from nominal wage trends. It is imperative to estimate real wages and skill premia in order to measure the different impacts of more fundamental changes on these three groups of labor. It is very useful to know exactly how much the monetary wages were for these three categories of labor. To do so, a benchmark wage needs to be determined. There were no widely accepted census wage data that I can use as a benchmark nominal wage. So I have to generate benchmark wage from this dataset. Ideally I want to choose a benchmark group that observations in this job category for this period are relatively homogenous, that is, these observations pertain to roughly same regions, with similar tenures and occupations are close. I choose unskilled wages from 1876 to 1878 as such a benchmark wage. There are seven observations in this group. Six are seamen, and the other is a lightkeeper. They are all from southeast China, and the tenures range from 1 year to 3 years. Relatively homogenous observations in this group allow me to do a simple average of the wages of these seven observations, and the result is 15.18 silver dollars. This is used as the benchmark nominal wage. The coefficients obtained from the hedonic regression stands for the log wage differences between the current year and the base year. To generate the nominal wage series, I recover this log differences, and use the benchmark nominal wages to generate wages for all three job categories in all the periods. The formula is: wagek ,t = wage1,1 ⋅ exp(δ k ,t ) (3)

where k=1,2,3 indicates unskilled, semi-skilled and skilled groups, and t=1,...,21 indicates totally 21 periods. The nominal wage series are reported in Table 4.

18

5. Cost of Living Indices Although nominal wages are useful, real wage estimates provide a richer picture of change over time in living standards. Reliable consumer price indices are required for the estimation of real wage series. Construction of consumer price indices requires price series of representative commodities and their weights in consumption budgets. However, both of these two tasks are formidable given the changing taxonomies of commodities in the CMC’s trade records over long span of time, tens of different quantity measurements and currencies and their exchange rates, different types of trade that might have caused different valuation of commodities, lack of data on nontradable goods, our limited knowledge of how the composition of expenditures varied over household characteristics and over time, and the incidence of sharp changes or missing data in the trade records. Thus, my analysis is far from being perfectly precise. There are some existing price indices available. However, all these indices suffered from some critical limitations, preventing them from being used in my estimation of real wage series. 25 The cost of living indices developed in this section remedy the disadvantages of these existent indices. The price data are collected from the CMC’s trade statistics; therefore covering all the treaty ports from 1860 to 1936. The sample commodities are all consumer goods, and their weights were collected from the CMC’s family budge surveys. Therefore it is a long-run weighted national consumer price index.

25

The Nankai Index is probably the only long-run price index covering the period from the 1860s to 1949 (Hsiao 1974). It is obtained from the CMC’s trade statistics. But this index suffered from two problems: it is a simple average rather than a weighted average; and rather than a consumer price index it is a general index including a large number of industrial commodities. There are some cost of living indices available, which focus on consumption goods only. Prices and expenditure weights were collected from the market directly. However, it only covers prices in big cities such as Beijing, Shanghai, and Tianjin since the 1920s (Cai, Zhengya 1932) (Kong 1988).

19

Real inequalities came from both nominal inequalities and unbalanced impacts of price changes on different income groups. The poor and the rich usually have disparate consumption styles. A rise in bread price, for instance, will hurt the poor much heavily than the rich. Therefore real inequalities usually have more drastic swings than nominal inequalities. To take this factor into account, besides the construction of the general index, I estimate the expenditure shares separately for high, middle and low income groups. Applying different expenditure shares to the price series generates three cost of living indices for three income groups separately. This method contributes to a better measurement of real inequalities and skill premia.26 5.1 Basket of Representative Consumption Goods and Their Prices Totally 28 consumption goods are selected into this basket. Their prices are calculated from trade records in the CMC’s trade publications. This part of data is widely considered reliable. Unskilled and skilled nominal wages, which were calculated in the previous section, were adopted here to represent labor costs. Therefore my consumption basket comprises 30 commodities. The CMC’s trade publications provide us a regular, continuing source of trade data for commodities leaving and entering Chinese customhouses. The CMC’s records are believed to be “the only reliable and systematic material” (Chang 1982) to study modern China’s trade patterns and commodity prices. No other data source has such long time span and geographical coverage as the CMC’s publications do. The CMC collected and published quantities and values of all the commodities passing through each customhouse from 1858 to 1949. There are over 160 volumes of books for these publications. They have been computerized into a database. I select 28 consumption goods into my consumption basket. 26

A similar work can be found in Hoffman et al. (2002).

20

Because the data from different ports over such a long time are usually in different measurement and currency units, I convert them into same units, and then aggregate the port-level quantities and values into national totals. I divide the national aggregated values by the quantities, and obtain the invoice prices of each selected commodity by each trade type. Finally, following the adjustment schemes proposed by Cheng (Cheng 1956), I convert the invoice prices into market wholesale prices. The details are reported in A3. Due to the capability of this project, the consumption basket is much smaller compared to any existing price index in China or in other countries. Therefore care needs to be taken in the selection of representative commodities. I follow three guiding principles in the process of selection: representativeness, data availability, and time-consistency. Some commodities are crucial consumptions in household budgets, and they definitely need to be selected. Rice and wheat flour are two staples in Chinese diet. Cotton cloth, coal, kerosene, etc., are all commodities that almost every Chinese in that period consumed. Data of selected representative goods also need to be available. If certain representative goods are not available, proxies need to be carefully selected. For example, sugar is selected to represent all the seasonings. These selected commodities need to be time-consistent. As the indices cover almost 80 years, the quality and taxonomy of these commodities usually changed. Therefore selection of a relatively homogenous commodity helps control this bias. For example, shirting is selected to represent other cotton cloth. Sugar is also such an example. Incidentally, data for certain years of each commodity might be missing from trade publications. In this case, I use two other sources to supplement the data. One supplementary data source is (Hsiao 1974). The other source is price series recorded in the appendices of the Decennial Reports. The data recorded in these two sources are also

21

cleaned from the CMC’s trade publications, and their reliability is widely recognized. 27 If there are still data missing, I have to interpolate. Since such cases are not frequent, interpolation does little harm to the precision of the price indices. The cleaned price series are annual series. The annual average series smooth out the seasonality of these commodities. Furthermore, besides the annual series, I group them into three-year average price series because of two reasons: the price series are sometimes too volatile, and the nominal wage series are three year groups. In this way, I am able to obtain real wage estimation with both three-year average nominal wages and price indices. These commodities can be grouped into five categories: food, clothing, fuel and lighting, housing and sundry. The first three categories are easy to measure, because all of them are tradable goods, and my price series are highly representative in measuring the price changes in these three categories. Food consumption is further divided into cereals, meat, vegetables, fruits and condiments. In my dataset, rice, millet, and wheat flour represent cereals; meat (a collection of pork, beef and mutton), fish and egg represent all meat consumption; vegetables are represented by turnip, bean and other vegetables; fruits are represented by pears and persimmons; condiments are represented by bean oil, sugar and salt. The clothing consumption is represented by native cloth (nankeen cloth), mechanized cotton cloth (shirting), and silk products. The fuel is represented by coal, and the lighting item is largely kerosene. The other two consumption categories are largely composed of nontradable goods, whose precise price series goods are virtually impossible to obtain. There were neither detailed records of house prices nor apartment rents in modern China. I use two methods to estimate housing price index. One method is to put together scattered records of housing 27

But because they only cover a much smaller collection of commodities in a limited amount of time, I cannot directly construct price series from these two sources, and have to go back to the original trade records.

22

expenses from the CMC’s Decennial Reports, which was one of very few sources providing such information. The second method is to construct housing price series from the geometric average of labor costs and construction material costs. 28 The more precise method calculating the geometric average generates similar trends suggested by the scattered information in the Reports. Sundry consumptions are very complex. Expenses such as education, communication, sanitation, health, entertainment, religion and taxes all fall into sundry consumptions, and their prices are hardly to obtain directly. The way to estimate sundry consumptions is the following: firstly, a large part of sundries is service, and the nominal unskilled and skilled wages developed from the previous section are used to reflect that part; other parts of sundry consumptions are represented by tobacco, liquor, green tea, blanket and carpet, chinaware, paper, and clock.

5.2 Weights In the estimation of historical price indices, usually the scanty information of family budgets does not allow the applications of either Laspeyres or Paasche method. Often times the way to aggregate the constituent price series is to generate an expenditure weight for each of the constituent commodity. The consumption behavior changes all the time, and ideally we want to use different weights for different periods. In this way the substitution effect is effectively controlled. But usually there is no enough information to generate weights for different periods. My commodity weights are developed from both family budget surveys conducted by sociologists and local government and the CMC’s surveys on cost of living expenses. In 28

This method was used by David and Solar. Appendix explains how the rent estimation is generated in this index.

23

modern China sociologists and local governments started to survey family budgets and living conditions of residents in major cities in the early twentieth century, and their results are generally considered to be reliable and have been widely used. A list of these surveys can be found in Appendix.29 Most of these surveys focus on family budgets of low-income residents, and only very few surveys cover high-income groups such as merchants, government officials and intellectuals. Another conspicuous limitation of these surveys is that they only focus on biggest cities, and it is not clear how the consumption behaviors vary between big and small cities. The geographic limit of these well-known surveys is remedied by the exploitation of newly accessible CMC surveys on cost of living expenses in all the customhouses around the nation. As mentioned in Section 3, CMC started to collect information about cost of living expenses and family budgets in all the treaty ports for the purpose of setting appropriate salary standard. Nearly 50 port-cities have been surveyed, so these archives are able to disclose more detailed information about geographical variation in consumers’ behaviors. One disadvantage of weights in modern China is that most surveys were conducted in about same periods. This does not allow scholars to check whether the consumer behavior had changed significantly over time. The CMC’s surveys on cost of living indices were conducted twice: once in 1920, and the others were conducted from 1929 to 1931. By comparing the differences between these surveys, we can the weights have changed in ten years. The CMC’s survey in 1920 reports expenditure shares of five budget categories for about 20 low and middle CMC employees. Very little information about shares on specific commodities is reported. I group these 20 occupations into unskilled and skilled categories 29

Some of these surveys targeted suburban peasants, so they are not included in my study.

24

using the same rule described before. No high-rank employee was surveyed. I aggregate local survey results into a national average, and the result is reported in Table 5. The CMC’s six surveys from 1929 to 1931 report detailed expenditure information on low income employees. Each customhouse only reported one single survey form, summarizing the household budget situation of average low-rank CMC employees. Therefore no information on middle or high income employees was surveyed. But one distinctive advantage of these six surveys is that expenses on about 40 major consumer goods were surveyed and reported, allowing scholars to study the weights of each specific commodity in greater detail. I group these commodities into five categories, and the result is also reported in Table 5. Despite these distinctive advantages of the CMC’s surveys, it also suffers from two problems: first, there is no information about living expenses of high-end consumers; second, all survey subjects were the CMC’s own employees. It is helpful to know the consumption patterns of people working in local private sectors. For this purpose, I supplement the analysis of the CMC’s surveys by those well-known surveys conducted by sociologists and local governments. The formats, the contents and the methods of these surveys were quite different. The way of aggregating these surveys is similar: I group the results into five basic consumption categories and subcategories. Then I take a simple average of these surveys to generate an average weighting scheme. This result is also reported in Table 5. Table 5 shows that there is no enough evidence to believe that consumption patterns changed in the ten years from 1920 to 1929-31. This suggests that life styles of common Chinese did not change in a short time, and using a single weighting scheme does not incur serious bias. Table 5 also suggests that the Engle’s law holds in modern China. Poor people had larger shares of expenses on food, while rich people cared more about sundry

25

expenditures such as education, entertainment, medicine. Housing expenses are relatively small for all income levels, ranging from 8.71 to 13.89 percent.30 As the information from both the CMC’s surveys and the surveys conducted by sociologists and local governments does not show significant difference, using either source does not lead to different conclusion in the construction of price indices. The weights used in this paper are developed from the surveys by sociologists and local governments. Since the current consumption basket contains 30 commodities, the weight of each commodity needs to be estimated. However, those surveys contain much more representative commodities, and the way to estimate the weights for my basket from those surveys is the following: the total expenditure is categorized into five groups: food, clothing, rent, fuel and lighting, and sundry. The food consumption is further categorized into cereals, meat, vegetables, fruits, and condiments. The clothing consumption is further categorized into three groups: native cotton goods, mechanized cotton goods, and silk goods. The fuel and lighting consumption is further categorized into coal and kerosene. Sundry consumptions are categorized into service costs and other costs. Rents do not need to be further categorized. The weight of each subcategory is estimated from those surveys. If necessary, the subcategory consumption can be further categorized. For example, cereals are categorized into rice, millet and sorghum, and flour. For each existing survey, I calculate the weights of categories, subcategories, sub-subcategories, and so on, and apply the average weights to my basket. For the lowest layer, for example, fruits under food category, or other sundry items under sundry category, sometimes weights are not able to determine. In this

30

The housing expenses in modern China seemed to be small, but it is not possible. The housing expenses in the US in the nineteenth century were estimated between 13 to 18 percent, which was also not high. For details, see David and Solar, Sokoloff and Villaflor.

26

case, I assign equal weights to them. Since this is the lowest layer, this will not lead to a significantly different result. The cleaned weights are reported in Table 6.

5.3 Cost of Living Indices By applying the weights to the price series, I construct the national average cost of living index and the indices of different income group. Figure 5 reports these cost of living indices. As shown in this result, the weights do not differ significantly over different income groups. The slight difference between the low and high income groups is that the cost of living index for low income group is more volatile. During deflation period, low-income CPI dropped more; during inflation period, low-income CPI rose more. Figure 6 compares my CPI with the existing Nankai index and the cost of living index in Shanghai from 1912 to 1936. There is no distinctive difference between three indices.

6. Real Wages and Skill Premia With the new cost of living indices for different income groups, the nominal wage series can be converted to the real ones. Table 4 reports the real wage series, and it is convenient to compare the nominal and real wage series from this table. Figure 7 plots the log wage series for three income groups. This figure suggests that real wages for unskilled labor experienced a very slow decline from the early 1880s to the mid 1920s. But generally unskilled real wage was very stable. The skilled wage rose first, and then leveled off from around 1880 to the end of the nineteenth century. It began to grow again from the beginning of the twentieth century, and reached its apex around 1910. This apex was followed by a sharp decline until the mid 1920s. Highly skilled wage was much more stable. It grew very mildly until the late the 1920s. From the late 1920s to 1936, all three wage series started to grow very rapidly.

27

I divide skilled and highly skilled wages by unskilled wage and obtain two skill premia, as plotted in Figure 8. These two skill premia share some similarities but also differ in some aspects. The ratio of skilled over unskilled wage rose rapidly from 1858 to around 1880, followed by a slower rise until around 1910. The period from 1910 to the mid 1920 witnessed a sharp decline of this skill premium, and then this skill premium leveled off. For the wage ratio between highly skilled and unskilled labor, it grew all the way through the mid 1920s, followed by a sharp decline to 1936.

7. Educational Boom in Modern China The rise and fall of skill premia reflect the changing demand and supply of the labor market, in particular, that of the skilled labor, because the unskilled wage was stable over time. After industrialization took place in the late 19th century, the demand for unskilled labor increased dramatically, but the unskilled wages failed to increase because the enormous repository of cheap labor in rural China met all the unskilled labor demand. Improved transportation gave cheap rural laborers much easier access to the cities, and resulting hotter competitions for unskilled jobs kept the unskilled wages at a very low level. The rapidly growing industrial sectors also created demand for skilled labor. In particular, the types of skilled labor in great need by modern industries were completely different from the traditionally defined skilled labor. They needed to be equipped with knowledge on engineering and mechanics, foreign language, and business and management, etc. Since the indigenous education failed to train students with such skills, the very scarce supply and increasing demand of skilled labor drove up the skilled wages as well as the skill premium. The shortage of new skilled labor was quickly responded from the supply side. China’s education system was fundamentally reformed in first three decades of the twentieth century

28

to adapt to the new politico-economic environment. In 1904, the central government issued edicts to establish a national school system, and in 1905 it abolished the Imperial Civil Service Examinations. This drastic reform was continued and further strengthened in the Republic period. The organization of the new education system, implemented first in 1904 and was continually evolving in the following two decades, was disparate to the indigenous education system in pre-modern China. In the traditional system, despite the existence of some official schools and academies, the government was content to leave most education in the hands of local communities and households. In sharp comparison to that, the government played a critical role in the design and implementation of the education reform. The national school system established in 1904 was to embrace six distinct levels of modern schooling: lower primary (5 years), upper primary (4 years), secondary (4 years), higher schools (3 years), and university (3-4 years). In 1915 the compulsory four-year education was legalized by the Republican government. The subjects taught in the new system fundamentally distinguished itself from its predecessor. Aiming at successes in the Civil Service Examination, the traditional education comprised a very stable set of knowledge categories: Confucius classics, history, philosophy and literature. Practical fields of knowledge relating to medicine, agriculture and engineering were regarded more as crafts than as knowledge and their development was overseen by scholar-officials who had the all important credentials in classical knowledge conferred by the examination system. Nothing about science and technology was ever taught. The traditional mass education, characterized by cripplingly burdensome memorization of unrelated individual characters, failed to provide any practical skill for students.

29

To meet the political demand of the building a strong and modernized nation and the economic demand of labor force equipped with knowledge of science and technology, foreign languages, business and management, law and so on, the curricula in new schools emulated those in Japan, Germany and America, marking a rupture from the past. Trainings of practical skills, particularly those on science and technology, were highly prioritized. The subjects taught in secondary schools included Chinese language, foreign language, mathematics, history, physics, chemistry, law, economics, geography, and so on. High education was classified into following specializations: arts, science, law, business, medical, agriculture and engineering. Even more than adding practical subjects into general education, the new system built a separate vocational education, specially focusing on the trainings of practical skills required in the rising modern industry. The purpose of vocational education was “to promote agriculture, industry and commerce etc in order to richen the country and its people”, and the principle was “to focus on practical skills instead of inane theories”.31 Did surges of education reform in the early twentieth century provide enough skilled labor to drive down the skilled wages? To answer this question, the sizes of demand and supply of skilled labor need to be estimated. An estimation on the size and distribution of workers working in modern industrial sectors shows that the total size of workers was about 110,000, mainly distributed in silk, ship-building, military, cotton textile and mining industries at the end of the nineteenth century.32 Modern industry started to grow rapidly in the first decade of the twentieth century. It was estimated by various scholars that the total population of modern workers ranged from 600,000 to 900,000.33 The second decade of the

31

The Emperor’s Edict on guiding principles of setting up vocational schools, 1903 Sun 1957, p1202; Tang et al [2007], p31 33 “First Labor Yearbook” p13; Wang p ; Tang et al [2007], p52 32

30

twentieth century witnessed the golden period of the development of modern industries. The size of workers quickly expanded to 2,885,000.34 Statistics on the education development in modern China was able to demonstrate that the size of modern education was big enough to meet the demand. Given that skilled and managerial employees usually received at least high school education, the size of secondary and higher education mattered. Table 4 shows that the number of enrolled students in total and in different types of secondary and high educations. There was no detailed statistics about the number of graduates from these schools. However, it is reasonable to say that this scale of education was sufficient to drive down the skilled wages.

8. Conclusion Although there was virtually no GDP or other direct estimate of the macro economy, numerous evidences suggest that China’s coastal regions, if not the entire country, realized rapid industrial development from the last decade of the nineteenth century to 1936, particularly in the 1920s, although quite often disturbed by political instabilities. However it remains unclear how the rapid industrial development affected real income and living standards of people in those areas. Did everyone benefit, or did different groups of people benefit differently? This paper shows that, real wages for all income groups began to grow rapidly since the late 1920s. Before that, unskilled real wages declined very mildly; skilled wages grew rapidly until an abrupt decline after around 1910; and the highly skilled wages remained very high and stable. With the real wage series of different labor groups, this paper estimates the changes in skill premia over time, and found that skill premia rose rapidly during the first two decades

34

Tang et al [2007], p73

31

of industrialization. After the 1910s, the wage gap between skilled and unskilled labor began to decline, while the gap between highly skilled and unskilled labor leveled off. These changes in the skill premia, I show, were driven by movements in the wages of skilled and highly skilled labor. China’s enormous reservoir of unskilled labor kept unskilled wages flat throughout the period. By contrast, supplies of skilled and highly skilled labor were extremely scarce, driving up the skill premia in the first few decades of industrialization. After the 1910s, however, progress in education increased the supply of skilled labor. Skilled wages started to decline, but wages for highly skilled labor stayed relatively high because the supply of this type of labor remained insufficient. Real wage series and changes in skill premia in modern China suggest that a deskilling technological revolution does not necessarily lead to a narrowing wage inequality. In developing countries with large stock of unskilled labor and small amount of skilled labor, such as China, the introduction of unskilled-labor intensive technologies will still possibly lead to an increasing wage inequality. An effective way of narrowing wage inequality, as suggested by this paper, might be to increase investments in massive education.

Primary Archival Sources “China Maritime Custom Archives”, the Second Historial Archive of China, Nanjing, China “Returns of Trade and Trade Reports”, the Imperial Maritime Customs of China, 1859 to 1936, “Decennial reports on the trade, navigation, industries, etc., of the ports open to foreign commerce in China and Corea, and on the condition and development of the treaty port provinces, the Imperial Maritime Customs of China, 1881 to 1931 Maritime Custom Project Website: http://www.bris.ac.uk/history/customs/about.html

References (Incomplete) Acemoglu, Daron. "Directed Technical Changes." NBER Working Paper No. 8287. Acemoglu, Daron. "Labor- and Capital-augmenting Technical Change." NBER Working Paper No. 7544, 1999. Acemoglu, Daron. "Patterns of Skill Premia." NBER Working Pape No. 7544, 1999.

32

Acemoglu, Daron. "Technical Change, Inequality and the Labor Market." Journal of Economic Literature XL (2002): 7-72. Acemoglu, Daron. "Why Do New Technologies Complement Skills? Directed technical Change and Wage Inequality." Quarterly Journal of Economics 113, no. November (1998): 10551089. Autor, David, and Lawrence F. Katz. "Trends in U.S. Wage Inequality:Revising the Revisionists." mimeo, 2007. Bickers, Robert, and Hans van de Ven. China and the West: The Maritime Customs Service Archive from the Second Historical Archives of China, Nanjing. London: Thomson Gale, 2007. Brady, Dorothy S. "Consumption and the Style of Life." In American Economic Growth, An Economist's History of the United States, by Lance E. Davis et al., 61-89. New York: Harper & Row, 1972. Cai, Zhengya. The Cost of Living Index Numbers of Laborers, Greater Shanghai, January 1926 December 1931. Shanghai: Bureau of Social Affairs, The City Government of Greater Shanghai, 1932. Caselli, Francesco. "Technological Revolutios." American Economic Review, 1999. Chang, Chung-li. The Income of the Chinese Gentry. Seattle, Washignton: University of Washington Press, 1962. Chang, John. Industrial Development in Pre-Communist China. Chicago: Aldine Publishing Co., 1969. Cheng, Yu-Kwei. Foreign Trade and Industrial Development of China. The University Press of Washington, D. C., 1956. Chinese Maritime Customs Service Project. http://www.bris.ac.uk/history/customs/ (accessed September 17, 2007). David, Paul A., and Peter Solar. A Bicentenary Contribution to the History of the Cost of Living in America. Vol. 2, in Research in Economic History, by Paul Uselding, 1-80. Greenwich, Connecticut: JAI Press Inc. , 1977. Dernberger, Robert F. "The Role of the Foreigner in China's Economic Development, 18401949." In China's Modern Economy in Historical Perspective, by Dwight H. Perkins, 19-47. Stanford, California: Stanford Univeristy Press, 1975. Doepke, Matthias and Dirk Krueger. "Origins and Consequences of Child Labor Restrictions: A Macroeconomic Perspective." mimeo, 2006. Goldin, Claudia, and Lawrence F. Katz. "Technology, Skill, and the Wage Structure: Insights from the Past." American Economic Review, May 1996: 252-257. Goldin, Claudia, and Lawrence F. Katz. "The Origins of Technology-Skill Complementarity." The Quarterly Journal of Economics, August 1998: 693-732. Goldin, Claudia, and Robert A. Margo. "The Great Compression: The Wage Structure in the United States at Mid-Century." The Quarterly Journal of Economics, February 1992: 1-34. Hoffman, Philip T., David Jacks, Patricia A. Levin, and Peter H. Lindert. "Real Inequality in Europe since 1500." Journal of Economic History 62, no. 2 (2002): 381-413. Hou, Chi-ming. Foreign Investment and Economic Development in China, 1840-1937. Cambridge, Massachusetts: Harvard University Press, 1965. Hsiao, Liang-lin. China’s Foreign Trade Statistics, 1864-1949. Cambridge, Massachusetts: Harvard University Press, 1974. KongMin. Collections of Nankai Economic Index Materials (Nankai Jingji Zhishu Ziliao Huibian). Beijing: China Social Science Press, 1988. Liu, Ta-chung, and Kung-chia Yeh. The Economy of the Chinese Mainland: National Income and Economic Development, 1933-1959. Princeton, New Jersey: Princeton University Press, 1965. 33

Lyons, Thomas P. China Maritime Customs and China's Trade Statistics, 1859-1948. Trumansburg, New York: Willow Creek Press, 2003. Ma, Debin. "Modern Economic Growth in the Lower Yangzi in 1911-1937: a Quantitative, Historical and Institutional Analysis." Working Paper, 2006. Margo, Robert A. Wages and Labor Markets in the United States, 1820-1860. Chicago: The University of Chicago Press, 2000. Perkins, Dwight H. Agricultural Development in China 1368-1968. Edinburgh: Edinburgh University Press, 1969. Rawski, Thomas G. Economic Growth in Prewar China. California: University of California Press, 1989. Sokoloff, Kenneth L., and Georgia C. Villaflor. "The Market for Manufacturing Workers during Early Industralization, The American Northeast, 1820 to 1860." In Strategic Factors in Nineteenth Century American Economic History, A Volume to Honor Robert W. Fogel, by Claudia Goldin and Hugh Rockoff, 29-65. Chicago: The University of Chicago Press, 1992. Twitchett, Denis, and John K. Fairbank. The Cambridge History of China, Late Ch'ing, 1800-1911. Vol. 10. Cambridge: Cambridge University Press, 1978. van de Ven, Hans. "Notes from the field: The Maritime Customs Service Project." The Bulletin of the British Association for Chinese Studies, 2000.

34

Appendices A1: Regional Distribution of Cities Region

Number of Cities

List of Cities

Manchuria

12

Aihui; Andong; Dalian (Dairen); Harbin; Hunchun; Jilin; Longjingcun; Manzhouli; Shenyang (Moukden); Sanxing; Suifenhe; Dadongkou

North China

8

Beijing (Peking); Yantai (Chefoo); Qinhuangdao; Qingdao (Kiaochow); Longkou; Newzhuang; Tianjin (Tientsin); Weihaiwei

Lower Yangtze

7

Zhengjiang; Hangzhou; Nanjing; Ningbo; Shanghai; Suzhou; Wenzhou

Upper and Middle Yangtze

8

Changsha; Chongqing; Yichang; Jiujiang; Shashi; Wanxian; Wuhu; Yuezhou

Southeast Coast

9

Xiamen (Amoy); Guangzhou (Canton); Fuzhou; Jiangmen (Kongmoon); Kowoon; Lappa; Sanshui; Sanduao; Shantou (Swatow)

Guangxi and Hainan

6

Haikou; Qiongzhou; Longzhou; Nanning; Beihai (Pakhoi); Wuzhou

Southwest Frontier

4

Mengzi, Simao; Tengyue; Yadong

Taiwan

3

Tainan; Takow; Tamsui

Northwest

3

Gansu; Shaanxi; Shanxi

35

A2: Categorizations of Occupations Categories

Subcategories 1

Unskilled

2

3

4

Semi-skilled

5

6

7

Skilled 8

9

Unit: silver dollar Occupations

Freq.

Percent

Median Pay

nightwatchman

324

0.73

11.22

coolie

665

1.49

11.42

gatekeeper

93

0.21

11.42

soldier

263

0.59

12.24

boat crew

70

0.16

12.75

boatman

420

0.94

12.75

sampanman

96

0.22

14

seaman

327

0.73

14.28

lightkeeper

3,115

6.98

9.91

cook

110

0.25

12.24

office assistant

86

0.19

12.65

office boy

277

0.62

16.32

fireman

151

0.34

18.36

weigher

296

0.66

18.41

carpenter

267

0.6

19.74

boatswain

133

0.3

21.81

boat captain

164

0.37

25.21

watcher

778

1.74

27.53

examiner

202

0.45

54.53

postal clerk

5,836

13.09

28.33

teacher

99

0.22

28.33

office secretary

5,793

12.99

31.16

pilot

46

0.1

39.66

engineer

247

0.55

40.21

shroff

125

0.28

42.49

stenographer

3,610

8.09

46.74

writer

3,431

7.69

46.74

copyist

36

0.08

47.42

linguist

212

0.48

52.09

harbor master clerk

498

1.12

49.57

tidewaiter

1,952

4.38

63.74

probationary clerk

248

0.56

70.82

revenue auditor

635

1.42

80.73

clerk

13,715

30.75

84.98

Total

44,600

100

36

A3: Commodity Price Series: Sources, Problems and Adjustments The sources of price data for tradable goods are a series of publications called “Returns of Trade and Trade Statistics” compiled by the CMC’s statistical department. For each year, quantities and values of each commodity in every treaty port were recorded in the publications. Totally there are more than 160 volumes of trade records, and the size of the database is incredibly huge. Step 1 of obtaining price series is to computerize more than two hundred hardcopies of the trade publications. This is an extremely time-consuming and costly project. Headed by Hans van de Ven and Robert Bickers, historians from University of Cambridge and University of Bristol and I together have spent more than three years in computerizing these publications. The database will eventually publish on the Internet for everybody’s use. Step 2 is to calculate the national average prices from trade publications. First, I add up port-level data for each selected commodity into national aggregates. I divide the national aggregated values by the quantities, and obtain the invoice prices of each selected commodity by each trade type. Trade types include foreign export, foreign import, domestic export, domestic import, and re-export. I focus on foreign exports and imports because of data availability. This step is very difficult to do, because measurement and currency units varied tremendously over locations and over time. The details of converting measurement and currency units are reported in A4. Step 3 is to convert invoice prices into wholesale prices. Before 1903 the CMC recorded the market values of commodities, so no adjustment is needed. Since 1904, the CMC began to report FOB and CIF values for exports and imports. The following equations are adopted to recover the market values from FOB and CIF values: CIF Value = (Market Value – Import Tariff) * (1 – Miscellaneous Fees) FOB Value = Market Value * (1 + Miscellaneous Fees) + Export Tariff Both import and export tariffs were 5%. The miscellaneous fees of imports and exports are estimated to be 7% and 8% respectively.35 Therefore the adjustment equations are: Market price of Imported Goods = CIF Value / 0.88 Market price of Exported Goods = FOB Value/ 1.13 The last step is to adjust the systematic bias in the CMC’s trade statistics. Most scholars agree that the CMC has consistently undervaluated the export values, but they have different estimations on the degree of this bias. 36 On the basis of existing studies, the following estimation scheme is adopted: export values were undervaluated by 5% before 1914; by 8% in 1914-1919; by 10% in 1920-1929; by 12% in 1930-1936. The export values and prices are adjusted accordingly. A4

Measurement and Currency Units For measurement unit, picul was the most often used weight unit in reporting trade volumes. The conversion rate of a picul and a ton is 1 ton = 16.535 piculs. Since 1930s quintal was used to replace picul, and 1 quintal equals 0.1 ton. 35

Quite a few scholars adopted 4% fees of both imports and exports. See, for example, G. Jamieson (1894), Zhou (1986), Yan and Wang (1998). However, since 1900s, more bulky commodities, such as coal, cotton, machinery, steel, and chemical products were traded, and miscellaneous fees were actually higher. Considering this, the CMC statistical reports estimated the fees to be 7% and 8% for import and export fees, and was adopted by various scholars such as Yao, Zheng (1984) an Hou. 36 See, the review article in Wang (1998) p 216 - 218

37

The currency problem was much more complex. Actually the chaotic currency system has been one of the biggest obstacles in the study of modern Chinese economy history. Copper cash was usually used in the small-amount transactions in people’s daily life. In early periods large-amount business transactions were carried out in silver taels. Later on silver taels were replaced gradually by silver dollars, which were initially imported from Britain, Spain and Latin American countries. The biggest problem for both silver taels and silver dollars is that there were many types of taels and dollars and their values varied. In the 1930s, the Republican government issued the “Fapi” (referred to as the Standard Dollar and later as the Chinese National Currency) as the only legal currency all over the country. Tremendously pestered by such chaos, the CMC invented its own fictitious currency, the Customs Tael (“HKT” hereafter), as the accounting unit, and uniformly adopted it in its own trade and wage records until the 1930s. Due to serious silver depreciation in the early twentieth century, the CMC decided to switch to the gold standard, and invented a goldbased currency, the Custom Gold Unit (“CGU” hereafter) to replace HKT in levying import duties in 1930 and in reporting foreign imports in 1933.37 What makes things even worse is that for treaty ports near the borders, foreign currencies were often adopted in accounting and recording. Reports from northeastern ports often adopted Russian ruble and Japanese yen, and Indian rupee was sometimes used in southwestern customhouses. A reliable exchange rate in that era was sometimes very difficult to find. The currency-conversion rates adopted in this project follows below. First I convert all the currency units into HKT using the rates below. But for the purpose of making comparison with prices from other sources, I then convert HKT values into silver dollar values. 1 HKT= 1.08 TLS 1.558 STD 0.845 CGU in 1932 0.798 CGU in 1933 0.792 CGU in 1934 0.835 CGU in 1935 0.689 CGU in 1936 A5 Sources and Notes of Price Series Food consumption: 1) Rice: rice price series are available from various sources. Hsiao (1974) recorded quantities and values of imported rice. The CMC also recorded average export and import rice prices in their decennial reports from 1862 to 1921. In this project, the long-run export rice prices from 1859 to 1936 are drawn from all CMC’s raw trade statistics. 2) Millet: millet and other grains such as sorghum are substitutes for rice, and were mainly consumed by the poor. The millet export data were also recorded in the trade statistics, their prices showed the similar trend with rice price. 3) Wheat flour: all prices are extracted from foreign import trades. The CMC collected prices of imported wheat flour and exported wheat in 1862/1921. This project calculates imported wheat flour prices from the raw trade data. 37

In foreign exports HKT was still used

38

4) Meat: the CMC recorded meat imports and exports, which includes a collection of pork, beef, mutton and etc. Here I use meat export prices. 5) Fish: the CMC’s trade records report dried and salted fish of all kinds. The fish price series here is generated from the export records of dried and salted fish. 6) Egg: the egg price series is generated from native export of preserved eggs, recorded in the CMC trade publications. 7) Beans: green, yellow and red beans were recorded under the category of bean in the CMC’s trade statistics. Beans are large items in the Chinese diets. They are either directly consumed, or made into bean curds and other products. The data source is the export of green beans from the CMC trade records. 8) Turnip: turnip, dried and salted, is one of the most widely consumed condiments in China. Its native export was recorded and the price was calculated here. 9) Vegetables: other vegetables together were recorded in the raw trade statistics. The native export data were recorded here, and the price was calculated. 10) Bean oil: bean oil is the most widely used cooking oil in China, supplemented by groundnut oil and sesame oil. The CMC’s Decennial Reports recorded the prices of bean oil and groundnut oil, and they show quite similar trend. Therefore using any of these oils does not affect the long-run trend. This project uses the bean oil price from the Decennial Reports, supplemented by prices from 1922 to 1936 calculated from raw trade data. 11) Sugar: sugar is one of the most important condiments in the Chinese diet. Sugar foreign import price here is calculated from trade statistics recorded in Hsiao (1974). Data from 1864 to 1866 are calculated from the original trade statistics. 12) Salt: salt used to be an expensive condiment in China. Its native export was recorded in the CMC’s records, and was used to calculate salt price. 13) Pear: pear is a common fruit in China. Its native export was recorded in the CMC’s records, and was used to calculate its price. 14) Persimmon: persimmon is also a common fruit in China. Its native export was recorded in the CMC’s records, and was used to calculate its price. Clothing consumption: 15) Shirting: shirting is used to represent all the mechanized cotton cloth. There were different types of imported cloth, such as grey shirting, white shirting, sheeting, drills, lastings, etc. Using shirting import data to calculate its price and represent other cotton cloth does not lead to a significant bias. 16) Nankeen cloth: many Chinese people, mostly with low income, consumed a large amount of native cloth. Nankeen cloth is a typical native cloth, and its price is calculated from the export record. 17) Silk: white silk was a luxury and was mostly consumed by the rich. China is a major silk-exporting country, and white silk export price is calculated from trade statistics. Fuel and lighting consumption: 18) Coal: coal is the most important fuel and heating material in China. Its export data were recorded in the trade statistics and the price is calculated accordingly. 19) Kerosene: kerosene was the most important lighting material, and was all imported. Its price is calculated from the raw trade data. Housing consumption It is common that housing costs were not available in economic history of many countries. To remedy this, I adopt the estimation method proposed by David and Solar. This method approximates rent by an index of the reproduction costs or prices of new structures. This index is calculated as a geometric average of indices for building materials prices and 39

common labor wages. This method corresponds to the dual of a constant return to scale Cobb-Douglas production function characterized by output elasticities of 0.5 for its labor and material inputs. My common labor wage series was developed from the CMC’s unskilled employees’ wage dataset. I extract prices of two building materials from the CMC’s trade publications: timber plank and cement, each is weighted by 0.25. An alternative way of estimating housing costs is to collect scattered or anecdotal records from historical materials. Among them, the CMC’s Decennial Reports provide relatively richer observations. However, all these observations are port-level. There is no practical way of aggregating national averaged rents from these fragmentary records. The impression from these observations is that the housing costs started to grow rapidly in the early twentieth century due to fast urbanization. Improved transportation condition lowered the cost of migration for rural labor, and industrialization provides more job opportunities for them. Booming population in cities led to a sharp increase in housing costs. This coincides with the picture generated by the first estimation method. The sources of price series for two building materials follow below: 20) Cement: foreign imported cement records are used to calculate prices. 21) Timber plank: native export records of timber plank were recorded in the CMC’s trade publications, and the price is calculated accordingly. Sundry consumption 22) Unskilled wages: unskilled wages are generated in Section 4. Unskilled labor represents many types of service costs in sundry category. 23) Skilled wages: skilled wages are also generated in Section 4. Skilled labor represents expenses on teachers, doctors, etc. 24) Green tea: green tea is widely consumed in China. Its export price is calculated from trade statistics recorded in Hsiao (1974). Data from 1864 to 1866 are calculated from the original trade statistics. 25) Tobacco: foreign imported prepared tobacco records are used to calculate prices. 26) Liquor: foreign imported liquor records are used to calculate prices. 27) Blanket and carpet: their foreign import records are used to calculate prices 28) Paper: paper represents consumptions on reading and writing. Its import records are used to calculate prices. 29) Chinaware: chinaware represents housing furnishing expenses, and its native export records are used to calculate prices. 30) Clock: clock import records are used to calculate prices.

40

A6 Sources of Family Budgets Period

Location

Author

Survey targets

Size

1918

Beijing

Dittmer

Tsinghua College employees

93

1923-1924

Beijing

Chen Ta

Tsinghua College employees

141

1922-1923

Beijing

Morrow

Factory workers

200

1922-1923

Beijing

Unknown

Factory workers

77

1925-1926

Beijing

L.K. Tao

Worker and teacher families

60

1926-1927

Beijing

Gamble

Worker families

283

1929

Shanghai

Ding and Zhou

Worker families

21

1927-1928

Shanghai

Yang Ximen

Families of workers in cotton spinning industries

230

1929

Shanghai

1934

Shanghai

1938

Chengdu

1939-1940

Chengdu

Cai Zhengya Worker families 305 Social Bureau, Shanghai Rickshaw coolies 304 Municipal Government Families of workers, small business owners and officials Yang Wei 588 and intellectuals Sun Huijun Families of various classes 27

1926-1927

Tianjin

Feng Huanian

Families of artisan workers

132

1928

Tianjin

Lin Songhe

Workers in chemical factories

177

1929-1930

Wuhan

1930

Railway workers

Chen Huayin Worker families 625 the Railway Minister of the Railway workers in Bejing-Shanghai Railways and 178 Central Government Shanghai-Hangzhou-Ningbo Railway

41

Table 1 Summary Statistics of Nominal Wage Unit: silver dollar N

Mean Salary

Standard deviation

Unskilled

5,846

12.63

7.37

Skilled

1,990

28.35

21.42

Highly skilled

36,483

64.99

50.79

Senior Officials

280

256.26

124.62

1858-1875

736

55.29

70.93

1876-1889

4,230

39.2

35.55

1890-1899

4,504

34.27

31.18

1900-1909

12,834

42.6

33.59

1910-1919

10,427

63.58

51.55

1920-1929

11,094

82.29

65.84

1930-1936

775

115.5

88.33

Manchuria

1,770

73.9

46.81

North China

6,427

58.4

53.9

Lower Yangtze

13,044

59.1

49.08

Upper and Middle Yangtze

7,045

49.01

46.14

Southeast Coast

12,613

60.49

61.2

Guangxi and Hainan

2,081

56.75

55.04

Southwest

1,238

45.14

39.7

Taiwan

219

43.56

31.95

Northwest

163

25.33

18.04

44,600

57.69

53.2

Job Categories

Periods

Regions

Total

42

Table 2

Means of Monthly Wages Recorded in CMC’s 1929-1931 Surveys CMC

Chinese firms

Foreign firms

29-Mar 29-Sep 30-Mar

11.76 12.16 13.08

12.22 12.34 12.76

13.24 13.49 14.19

30-Sep 31-Mar 31-Sep

13.2 14.08 14.11

12.6 13.5 13.56

14.4 15.52 15.23

Dates

Regions Manchuria North China Lower Yangtze Upper and Middle Yangtze Southeast Coast Guangxi Hainan Taiwan Southwest Occupations boatman boatswain bricklayer cabin hand carpenter coolie coxswain

15.91 11.08 12.70 11.59 15.78 12.12 7.82

16.36 10.13 11.83 10.62 17.08 11.96 5.19

19.98 11.30 12.47 12.44 17.74 16.01 7.80

10.20 26.78 20.06 11.93 15.74 10.43 18.75

8.33 19.26 25.03 15.19 15.95 9.05 33.39

9.13

deck hand engineer fireman gardener gatekeeper hulkkeeper junk boat owner mason mechanic messenger motorman office boy sailor seaman watchman

12.34 24.98 15.13 10.12 10.27 11.16 20.15 14.46 27.58 9.58 22.27 11.12 15.26 15.20 10.30

12.12 23.22 13.31 11.11 8.19 11.24 20.19 15.26 25.87 8.35 8.86 11.00 8.71

12.18 18.45 10.43 22.77 12.37 23.35 14.41 10.59 9.85 17.09 20.82 17.92 27.04 10.23 18.61 12.25 10.91 14.00 10.96

Source and Note: China Maritime Custom Archives No. 679(1) 16233-16238; Wages are in silver dollars

43

Table 3 The Logarithms of Wage Differences, Surveys 1929-31 unskilled

skilled

highly skilled

0

0.274***

0.787***

[0.016]

[0.021]

-0.222***

0.161***

0.711***

[0.014]

[0.024]

[0.031]

0.0161

0.3397**

0.7752

[0.013]

[0.024]

[0.029]

CMC Chinese firms foreign firms

Notes:

Occupation dummies are not reported. Omitted groups are logarithms of monthly wages from CMC, in Manchuria and of the unskilled.

Table 4 Nominal and Real Wage Series (in silver dollar) Nominal Wages year 1858-1875 1876-1878 1879-1881 1882-1884 1885-1887 1888-1890 1891-1893 1894-1896 1897-1899 1900-1902 1903-1905 1906-1908 1909-1911 1912-1914 1915-1917 1918-1920 1921-1923 1924-1926 1927-1929 1930-1932 1933-1936

unskilled

skilled

17.14 15.18 14.03 13.60 12.63 11.76 12.23 12.29 13.51 15.51 16.00 15.80 21.15 16.25 21.12 18.97 24.02 36.15

30.52 32.08 40.38 37.74 35.78 36.53 36.70 35.13 42.23 43.74 48.22 50.76 70.34 45.54 40.08 37.70 29.81 40.54 66.54

61.79

114.37

Real Wages highly skilled 64.65 59.98 59.30 65.91 64.42 62.23 67.14 61.71 69.87 80.74 84.81 87.42 106.79 87.02 89.55 88.81 106.17 122.80 144.38 119.98 275.40

unskilled

skilled

17.14 16.66 16.38 16.81 15.16 14.41 15.04 13.22 12.43 13.88 12.55 12.54 15.14 11.94 12.72 11.31 12.85 17.59

28.52 33.36 44.32 43.06 40.93 42.36 43.08 36.20 38.64 38.55 40.48 43.11 54.29 36.96 31.98 25.39 20.04 24.83 34.91

29.87

55.48

highly skilled 58.10 60.24 62.87 72.81 72.84 70.53 76.41 63.53 65.24 71.63 74.25 77.44 85.87 75.08 75.18 63.85 75.29 80.81 78.98 57.03 134.67

Note: Estimations of cross terms of years and job categories are plotted in Figure 3, but not reported in this table due to space limitations. Estimations of controlling variables “proportion” are not reported here either.

44

Table 5 Family Budgets from Different Sources of Surveys sources

surveys

the CMC's surveys low income in middle income 1920 survey in 1920 survey

the surveys conducted by sociologists and local governments

low and middle income in 1929-31 surveys

low income

middle income

high income

average

7.58

27.89

46.67

12.52

average wage (in silver dollar)

9.85

30.51

food (%)

56.67

42.67

50.76

55.89

37.49

26.24

50.48

clothing (%)

12.96

15.34

12.70

8.08

8.80

10.18

8.39

rent (%)

13.59

13.89

12.85

8.71

12.70

13.06

9.67

fuel and lighting (%)

10.45

9.86

11.26

9.44

7.48

6.15

8.85

sundry (%)

6.33

18.24

12.43

18.01

33.86

44.37

22.75

45

Table 6 Weights of Representative Commodities average wage (in silver dollar) Rice Millet Wheat flour Bean oil Meat Egg Fish Turnip Other vegetables Bean Pear Persimmon Sugar Salt Native cloth Shirting Silk Rent Coal Kerosene Chinaware Paperware Clock Liquor Blanket and carpet Green tea Tobacco Unskilled wage Skilled wage Total

Low Income

Middle Income

High Income

Average

7.58 16.21 5.40 8.88 6.42 9.66 0.21 0.41 0.68 0.68 5.35 0.11 0.11 0.41 1.22 3.29 3.29 1.51 8.71 3.48 5.97 0.86 0.86 0.86 0.86 0.86 0.86 0.86 6.00 6.00 100.00

27.89 5.00 1.67 7.20 5.11 11.58 1.47 1.47 0.24 0.24 1.78 0.05 0.05 0.33 0.98 2.67 2.67 3.46 12.70 3.76 3.72 1.61 1.61 1.61 1.61 1.61 1.61 1.61 11.29 11.29 100.00

46.67 2.63 0.88 3.04 2.85 7.13 2.14 4.09 0.11 0.11 0.86 0.89 0.89 0.18 0.55 1.83 1.83 6.53 13.06 4.68 1.36 2.11 2.11 2.11 2.11 2.11 2.11 2.11 14.79 14.79 100.00

12.52 11.06 3.69 8.10 5.56 11.21 1.35 2.45 0.50 0.50 3.91 0.35 0.35 0.38 1.14 2.98 2.98 2.43 9.67 4.22 4.43 1.08 1.08 1.08 1.08 1.08 1.08 1.08 7.58 7.58 100.00

46

Table 7 Education Statistics in Modern China

Year

Number of Schools

1902

Number of Students

Number of High School and College Students

6,912

6,053

1903

769

31,428

8,562

1904

4,476

99,475

14,262

1905

8,277

258,873

85,026

1906

23,862

545,338

63,679

1907

37,888

1,024,988

203,321

1908

47,995

1,300,739

113,321

1909

59,117

1,639,641

170,229

1910

42,696

1,284,965

1911

52,500

1912 1913

87,272

1914

Number of High School Students

Number of College Students

2,933,387

137,912

97,798

40,114

3,643,206

157,399

119,026

38,373

4,075,338

151,133

119,057

32,076

1915

122,280

4,294,251

190,223

164,981

25,242

1916

129,528

3,974,454

163,802

146,561

17,241

1921 1922

4,987,647 178,847

6,615,772

1923

9,000,733

1924

11,067,635

118,656

1925

222,302

1928

129,978

36,321

234,811

35,198

1929

214,572

9,252,222

370,145

341,022

29,123

1930

253,917

11,501,152

552,175

514,609

37,566

1931

262,992

12,301,611

581,015

536,848

44,167

1932

266,578

12,812,983

589,917

547,207

42,710

1933

262,328

12,985,735

602,256

559,320

42,936

1934

263,915

13,771,380

583,247

541,479

41,768

1935

294,724

15,694,589

584,390

543,262

41,128

1936

323,452

19,034,124

669,168

627,246

41,922

47

Figure 1 China Foreign Trade Statistics, 1868-1936 Unit: thousand U.S. Dollars 1800

1500

1200

900

600

300

0 1868

1876

1884

1892

1900

Net Import

1908

1916

1924

Net Export

1930

1936

Total Trade

Sources: 1. 2. 3.

Original trade data, L. Hsiao, China’s Foreign Trade Statistics, 1864-1949 (Cambridge, Mass., 1974), pp. 22-25. Exchange rate of Haikwan Tael to U.S. dollar, L. Hsiao, China’s Foreign Trade Statistics, 1864-1949 (Cambridge, Mass., 1974), pp. 190-193. 1 Haikwan Tael equals 1.558 dollars (Chinese silver dollar). Figure 2 Quantities of Foreign Imports and Exports of Cotton Yarn, 1910-1935 Unit: thousand piculs

3,000

2,500

2,000

1,500

1,000

500

0

1910

1915

1920

1925

Foreign Imports

1930

1935

Foreign Exports

Sources: Imperial Maritime Customs / Chinese Maritime Customs, Returns of Trade (Shanghai: Statistical Dept., Inspectorate General of Customs), Statistical Series No. 5 Annual Publication.

48

-1

0

1

2

3

Figure 3 Log Nominal Wages, 1858-1936

1873-1875

1882-1884

1891-1893

1900-1902

Unskilled

1909-1911

1918-1920

Skilled

1927-1929 1933-36

Highly Skilled

80 100

150

200

250

300

Figure 4 Skill Premia on Nominal Wages, 1858-1936

1858-1875

1882-1884

1891-1893

1900-1902

skilled/unskilled

1909-1911

1918-1920

1927-1929 1933-36

highly skilled/unskilled

49

50

100

150

200

250

Figure 5 Cost of Living Indices for Three Income Groups

1858-1875

1882-1884

1891-1893

1900-1902

Low

1909-1911

1918-1920

Middle

1927-1929 1933-36

High

100

120

140

160

180

Figure 6 Comparison of My CPI with Other Indices

1912

1915

1918

1921

1924

Nankai Index

1927 Shanghai CPI

50

1930

1933 My CPI

1936

-.5

0

.5

1

1.5

2

Figure 7 Log Real Wages, 1858-1936

1858-1875

1882-1884

1891-1893

1900-1902

Unskilled

1909-1911

1918-1920

Skilled

1927-1929 1933-36

Highly skilled

80 100

150

200

250

300

350

400

Figure 8 Skill Premia of Real Wages, 1858-1936

1858-1875

1882-1884

1891-1893

1900-1902

skilled/unskilled

1909-1911

1918-1920

1927-1929 1933-36

highly skilled/unskilled

51