Reform, Growth, and Inequality in China*


  • *

    The author is grateful to Nhumai Hohti for research assistance and to participants in the conference on “Growing Inequalities” organized by the Asian Economic Policy Review and held in Tokyo on September 29, 2007, and especially the two discussants of the paper, Li Shi and Yu Yongding.

†Correspondence: John Knight, Department of Economics, University of Oxford, Manor Road Building, Oxford OX1 3UQ, UK. Email:


This paper provides a survey of the literature on inequality in China – level, change, causes, and consequences. It attempts to answer six main questions. How much has inequality risen? What is its relation to poverty alleviation? What has happened to wealth inequality? What are the main dimensions of rising income inequality? The dimensions examined are: the rural–urban divide; urban labor market reform; regional divergence; rural–urban migration; and entrepreneurship, rent-seeking, and corruption. Was it inevitable that inequality should rise so much? Does it matter that inequality has risen? Income distribution in China is bound up with both economic reform and economic growth. This paper concludes by considering the countervailing forces that will determine the path of inequality in future years.

1. Introduction

In 1978, China had a centrally planned economy, low productivity, widespread poverty, and very low inequality. Such inequality as then existed was regional in character, reflecting the lack of factor mobility among regions. The last three decades have witnessed three great and closely related events: a gradualist transition from central planning towards markets, a remarkable rate of economic growth, and a dramatic rise in economic inequality.

Of course, it is not surprising that inequality rose from its initial low level, as the economy became marketized and as egalitarianism ceased to be a central plank of economic policy. But China now has an unequal society – not rivalling the extreme inequality found in (for example) Brazil or South Africa, but nevertheless worse than or on a par with other parts of developing Asia. For instance, a study by the Asian Development Bank (2007, figure 1) finds China's inequality to be the joint highest (with Nepal) among the 21 Asian countries that are compared.

There is a vast literature, both theoretical and empirical, on the relationships between economic development and economic inequality. Kuznets (1955) postulated an inverted U-shaped relationship: as mean income grows, so the inequality of income per capita first rises, eventually peaks, and then declines. The main theoretical reasons for expecting an initial rise are the growth in the importance of the modern, or urban, sector of the economy, with its higher income and greater inequality, and the emergence of inequality based on possession of financial, physical, and human capital. The main reasons for expecting an eventual decline are that the low-income rural sector becomes a minority, that market scarcity drives up the returns to relatively unskilled labor, and that the socioeconomic groups that emerge with economic development acquire the political power to press for redistribution and the creation of a welfare state. The cross-country evidence for the inverted U-shaped relationship is actually very weak (Anand & Kanbur, 1993a,b), and Kanbur (1997) has argued for country case studies as a better way of understanding the complex relationships between economic development and economic inequality.

China offers an excellent case study. However, the fact that economic reform and economic growth have gone hand in hand means that it is difficult to distinguish their effects on inequality. China's experience cannot, therefore, serve as a reliable guide for other poor countries, except perhaps those countries that are both poor and also undergoing transition to a market economy.

There are many dimensions of inequality besides household income per capita. Other contributors to economic welfare include education, health care, housing, social security, subjective well-being, and what Sen (1984) has called people's “capabilities” to be and to do things of intrinsic worth. In order to make this summary paper manageable, I concentrate on income inequality but refer occasionally to other important dimensions of inequality.

The six main questions I want to address are:

  • 1How much has income inequality risen?
  • 2What is its relation to poverty alleviation?
  • 3What has happened to wealth inequality?
  • 4What are the main dimensions of rising income inequality?
  • 5Was it inevitable that inequality should rise so much?
  • 6Does it matter that inequality has risen?

2. How Much Has Inequality Risen?

The annual national household income and expenditure survey conducted by the National Bureau of Statistics (NBS) is not available to researchers at disaggregated household and individual level. Measures of income inequality such as the Gini coefficient, when based on official statistics, therefore have to be based on province-level or percentile data. However, there is another source: a national household survey designed and conducted by the Institute of Economics, Chinese Academy of Social Sciences (CASS), in conjunction with foreign scholars including myself, with research hypotheses in mind. This has been conducted every 7 years: in 1988, 1995, and 2002. I draw heavily on the results of research conducted within this China Household Income Project (known as CHIP) and based on these surveys.

Because of the sharp administrative and economic divide between urban and rural China, and the need for different survey questionnaires, Table 1 presents three Gini coefficients for each year, corresponding to urban, rural, and total households. In 1988 the urban Gini was low by international standards (24%), the rural Gini was higher (33%), reflecting regional income disparities, and the national Gini was higher than either (40%), reflecting the high ratio of urban to rural income per capita. Over the 7 years 1988–1995, all three Gini coefficients rose; the national Gini was 45% or 47% in 1995. We show two estimates of the Ginis for 1995 and 2002, from the same data sources. The normally small differences between them do not have a ready explanation and serve as a warning against being overprecise. (Possibly it arises because the first set of results contains only the provinces that were included in all three surveys whereas the second set of results apparently uses all available provinces.) However, the trend was not continued over the next 7 years. In 2002, the urban Gini (31% or 32%) was 2 percentage points lower than in 1995, the rural Gini (37%) was 1 percentage point higher or 4 points lower, and the national Gini remained constant, at 45% or at 47%.

Table 1.  CASS surveys: Gini coefficients of disposable income per capita, urban, rural and national households, China, 1988, 1995 and 2002 (percentage)
  1. Source: Gustafsson et al. (2007b, table 2.2); Khan and Riskin (2007, tables 3.3, 3.6, 3.9).

1995 (Gustafsson)343647
1995 (Khan and Riskin)334245
2002 (Gustafsson)323647
2002 (Khan and Riskin)313845

However, the finding of constancy in the Gini after 1995 is open to question. Ravallion and Chen (2007), of the World Bank, having partial access to the NBS micro-level data, show the Gini continuing to rise in the years 1995–2001 (Table 2). This is what might be expected on the basis of casual observation. Over those years, all the Gini coefficients increased, by 4 percentage points in the urban case, 2 percentage points in the rural case, and 3 percentage points in the national case; their estimate of the national Gini in 2001 was 45%.

Table 2.  National Bureau of Statistics surveys: Gini coefficients for urban, rural and total households, China, 1988, 1995 and 2001 (percentage)
  1. Source: Ravallion and Chen (2007, p. 46).


The annual China Statistical Yearbook of the NBS provides information on income shares of household income quintiles in urban China, from which it is possible to explore whether inequality (as measured by the NBS surveys) is continuing to rise in the 2000s. Table 3 shows that it has done so. The share of the first three quintiles falls monotonically over the decade 1995–2005, and that of the highest quintile grows sharply. Similarly, there is a monotonic rise in the (necessarily approximate) Gini coefficient. The 5-year percentage increases in income per capita rise monotonically across the quintiles. There is no indication that inequality has stopped rising.

Table 3.  Income by household income quintile, urban China, 1995, 2000, 2005: percentage of total; 5-year percentage increase in income per capita and in Gini coefficient
Quintile12345Gini (%)
  1. Notes: We obtain quintile income shares by multiplying income per capita in each quintile by household size. The percentage increases in income per capita are expressed in nominal terms. The Gini coefficient, being calculated from quintile shares, is liable to be inaccurate but its trend is less likely to be biased.

  2. Source: National Bureau of Statistics (annual), various issues.

Percentage share:
199511.115.218.522.6 32.620
2000 9.714.218.123.0 35.024
2005 41.432
Percentage increase:
1995–200042525966 7420

It is possible to obtain equivalent information in rural China for household income quintiles, households being ranked by income per capita, but it requires us to infer quintile income by applying assumptions to the frequency distribution of incomes. Table 4 shows that the income shares of the quintiles were stable over the years between 1995 and 2005, and that the distribution worsened, and the Gini coefficient rose, between 2000 and 2005. Whereas Ravallion and Chen (2007) calculated the rural Gini to rise over the period 1995–2001, and Gustafsson et al. (2007b) found it to be stable but Khan and Riskin (2007) found it to fall over the period 1995–2002, Table 4 suggests that it barely changed in the late 1990s and rose in the early 2000s.

Table 4.  Income by household income quintile, rural China, 1995, 2000, 2005: percentage of total, and Gini coefficient
Quintile12345Gini (%)
  1. Notes: The household income quintiles are ranked by income per capita and their income shares are derived from frequency distributions of income per capita by making various assumptions about mean income within frequency bands. The Gini coefficient, being based on quintile shares, is liable to be inaccurate.

  2. Source: National Bureau of Statistics (annual), various issues.

Percentage share:

Whether income inequality has recently stopped rising in China is clearly an important question. Both fact and feeling are relevant. The public perception of urban inequality is that it continued to rise. Respondents in the 2002 urban survey were asked how inequality had changed in their city over the previous 5 years. Offered a choice of five answers, over 90% reported that it had widened, 48% saying “slightly”, and 43%“notably” (Gustafsson et al., 2007b, p. 80).

Why do the two data sources produce different answers? The main reason is that they have different concepts of income. The NBS definition excludes the imputed value of owner-occupied housing and undercounts subsidies, whereas the CASS definition includes housing-related income components and more fully captures subsidies. These components of income disproportionately accrue to urban households, which are relatively rich. This explains why CASS-based estimates of the national and urban Gini coefficients tend to be higher than NBS-based estimates. Moreover, the importance of these income components has declined in recent years. This explains why the CASS-based Gini did not rise between 1995 and 2002. A recalculation of the CASS data using the NBS definition does indeed produce an increase in the Gini between 1995 and 2002 of 1.6 percentage points; and a value in 2002 of 45%– equal to the NBS-based estimate for 2001 (Gustafsson et al., 2007b).

The withdrawal of subsidies has a misleading implication. The most important urban subsidy, the housing subsidy, was merely capitalized: the housing subsidy was reduced after 1995 because urban houses were sold to their occupants at prices well below the prices that would have prevailed had there been a housing market. Zhao and Ding (2007, tables 5.5, 5.7, and 5.8) show that, in 2002, the ratio of current market price to purchase price of houses averaged over 7 to 1, and that the net value of housing represented 64% of total wealth and contributed 67% to the inequality of net wealth per capita (the Gini coefficient being 48%) in urban China.

The various estimates of income discussed above are for disposable income (i.e. after taxes paid to and benefits received from central and local governments). In fact, net taxes do nothing to remedy factor income inequality and its growth, either between rural and urban China or within these two sectors. Khan and Riskin (2007) show that net taxes reduced rural income per capita (by 2.6% in 2002). However, although regressive in 1995 (accounting for 2% of total rural inequality as measured by the Gini coefficient), net taxes became slightly progressive (contributing −1%) in 2002. By contrast, subsidies exceeded taxes in urban China, despite the greater prosperity of urban-dwellers. Net subsidies (mainly housing) accounted for 11% of income per capita in 1995 but fell to 2% in 2002. The contribution of net subsidies to urban inequality was 16% of the total in 1995 but only 1% in 2002: from being strongly regressive, fiscal intervention became only mildly so. However, there remains great scope for a more progressive fiscal policy.

3. How Does the Rise in Inequality Relate to Poverty?

In the 20 years between 1981 and 2001, real household income per person increased on average by 5.8% per annum in urban China, 5.4% in rural China, and (with reweighting) 6.5% overall (Ravallion & Chen, 2007, table 1). On the one hand, this rapid growth of incomes had the power to reduce poverty dramatically. On the other hand, the rising inequality tended to increase poverty. There were two counteracting forces: the rise in mean income and the rise in dispersion of income around the mean. In fact, the former proved much more powerful.

Table 5, taken from Ravallion and Chen (2007), shows poverty (as defined by government) declining sharply over the years, whether we use the headcount measure or the poverty gap index. The great fall in rural poverty occurred in the early period of decollectivization and rural reform, from 1978 to 1985. However, rural poverty incidence and intensity continued to fall throughout, despite rising inequality. Urban poverty has always been low, reflecting the protected and privileged status of urban-dwellers. However, research based on a 1999 CASS urban household survey that examined the effects of the massive redundancy program in the state sector, commencing in the mid-1990s, shows that urban poverty became closely associated with unemployment or layoff. Individuals who were unemployed or xia gang (laid-off), or whose household heads suffered that status, had a predicted probability of falling into poverty three to six times higher on that account (Li, 2006).

Table 5.  National Bureau of Statistics surveys: measures of poverty for urban, rural and total households, 1981, 1988, 1995, and 2001 (percentage)
  1. Note: The headcount measure of poverty (HC) is defined as the percentage of households that fall below the officially set poverty line, and indicates the relative number of poor people. The poverty gap measure (PG) is defined as the ratio of income needed to get all poor people up to the poverty line to income as a whole, and provides an indication of the relative resources required to eradicate poverty. The poverty line is officially set at 850 yuan per person per year in 2002 prices for rural China, and at 1 200 yuan for urban China.

  2. Source: Ravallion and Chen (2007, tables 2–4).

19882.10.523.2 5.517.7 4.2
19950.90.220.4 5.714.7 4.1
20010.50.212.5 3.3 8.0 2.1

4. What has Happened to Wealth Inequality?

During the period of central planning, there was almost no personal wealth in China. Starting from scratch, economic reform brought with it, in just a few years, not only the rapid accumulation of wealth but also considerable inequality of wealth. China thus provides an excellent case study of the processes that generate inequality of wealth. We draw on the research of Zhao and Ding (2007), who used the CASS datasets.

Urban real (net) wealth grew at a rate of no less than 19% per annum between 1995 and 2002. In the latter year, urban wealth per capita stood at 4.7 times urban income per capita, and the equivalent rural figure was 3.9. The ratio of urban to rural wealth per capita was 3.6:1.

The Gini coefficient of wealth per capita in China as a whole in 2002 was 0.55 – considerably higher than that of income per capita. The rural coefficient had risen from 0.35 in 1995 to 0.40 in 2002. The main contribution to this inequality was housing (accounting for 49% of the total in 2002). Although the imputed value of land (held on long leases) contributed a further 20%, the policy of allocating village land equitably ensured that the inequality of land reflected spatial differences in land endowments rather than localized inequality. The Gini coefficient of urban wealth per capita was 0.50 in 1995 and 0.48 in 2002. The most important contribution came from housing (accounting for 68% of the inequality in 2002) and financial assets (24%). We see the great importance of housing both for accumulating wealth (urban housing wealth grew by 22% per annum over the 7-year period) and also for generating inequality of wealth. Urban-dwellers who acquired ownership of the houses that they had occupied (paying nominal rents) made huge capital gains. The inequality of wealth in turn feeds the inequality of income through the returns to wealth assets such as imputed rents of owner-occupied houses.

More generally, the acquisition or appropriation of state assets at below-market prices was a powerful disequalizing force, which central and local governments in some cases aided and abetted and in other cases failed to prevent. This is the point that Yu Yongding makes so forcefully in his comments below. The divergence of wealth was also assisted by the fact that the household saving rate rises sharply with income. For instance, the 1995 CASS survey showed the urban saving rate to rise monotonically from −2% in the lowest income per capita decile to 35% in the highest; the corresponding rural rates were 3% and 41% respectively (Shi, 2007).

5. What are the Main Dimensions of Rising Income Inequality?

There are several dimensions of rising inequality, and to measure their contributions requires various forms of decomposition analysis. Here I concentrate on five important ones.

5.1 The urban–rural divide

This was the title of the book which Lina Song and I published (Knight & Song, 1999). There has been a very large ratio of urban to rural income per capita throughout the Communist period. It narrowed during the period of decollectivization and rural reform 1978–1985, but since urban economic reform commenced in the mid-1980s, the ratio has grown and is greater today (at 3.20:1) than it has ever been (Table 6, which shows both the NBS and CASS results). The high ratio was understandable when urban and rural China were compartmentalized, but it has survived and even grown as the “invisible Great Wall” that separated rural and urban China has been partly dismantled and marketization has occurred.

Table 6.  National Bureau of Statistics (NBS) surveys: the ratio of urban to rural household income per capita (rural = 100)
  1. Source: National Bureau of Statistics (2004, p. 357). Gustafsson et al. (2007b, chapter 2).


The ratio of urban to rural disposable income per capita rose consistently in the CASS data as well: from 2.69:1 in 1988 to 3.08:1 in 1995, and to 3.13:1 in 2002. The CASS ratio is higher because the definition of income used is broader, but the difference narrows over time with the withdrawal of many subsidies. The upward trend has helped to increase the contribution that the urban–rural income ratio makes to national income inequality. The contribution rose from 37% in 1988 to 41% in 1995 and to a remarkable 46% in 2002, far higher than for other developing countries in which similar decompositions have been conducted (Gustafsson et al., 2007a, table 1.2). Nearly half of China's income inequality would vanish if mean income per capita in rural and urban China were equal.

The size of the rural–urban income disparity might be exaggerated if the cost of living is higher in urban areas. For instance, Ravallion and Chen (2007, table 1) report that the urban cost of living was 19% higher than the rural in 1981, and that this rose to 40% in 2001. Using the 2002 dataset, Sicular et al. (2005, table 1) show that the ratio of urban to rural household income per capita is reduced from 3.4 to 2.4 by introducing urban–rural purchasing power parity. However, the rural–urban contribution to national inequality is reduced by less than a third (from 47% to 33%) when incomes are corrected for spatial price differences (Sicular et al., 2005, table 3). Moreover, as Li Shi points out in his comment below, the inclusion in measured income of in-kind subsidies such as subsidized pension schemes and health care, which are received largely by urban residents, raises the ratio of urban to rural household income per capita to 4.3, so bringing it back to roughly the original figure of 3.4 when adjustment is also made for differences in urban and rural price levels. The in-kind subsidies of course raise overall inequality and the contribution to this total that is made by the urban–rural divide.

How is the rural–urban divide to be explained? During the planning period the Communist Party built an institutional framework that divided China into two compartments, separate in terms of administration, financing, and resources. The policy of “price scissors” kept down the price of food in relation to manufactures. In that way the peasants paid indirectly for urban industrialization. Part of the investable surplus was diverted to enable urban workers to enjoy a higher standard of living than their rural counterparts. Worker discontent, low morale, and threat to political stability are all aspects of the power – latent rather than overt in the case of China – which residentially concentrated, interacting workers appear to possess (Knight et al., 2006).

As the economy became more marketized, why were the returns to factors not equalized across space? Why did the rural–urban income gap not get competed away? The answer lies in the government control of migration. Migrant workers were permitted to meet the growing needs of the urban economy but urban workers continued to receive preference – in employment, wages, housing, social security, etc. – while migrants were treated as second-class citizens. The real wages of urban hukou residents have increased rapidly over the last two decades of urban economic reform, being shielded from the labor market competition of rural labor.

Will the rural–urban divide be eroded in future? There are reasons to expect a gradual erosion. First, there is no longer any need to extract an investable surplus from the agricultural sector. Second, the voice of rural people may be enhanced by the greater awareness of their relative position that better education, greater mobility, and improved communications can foster. Third, the increased marketization of the economy may itself reduce the degree of urban bias in government policies insofar as it accords greater weight to market-based allocation of resources and less weight to location-based allocation. However, perhaps of greatest importance for the future is the evolving policy on rural–urban migration and the possible emergence of labor shortages, which we examine below.

4.2 Urban labor market reform

Under central planning the work unit (danwei) serving as a mini welfare state, providing lifetime employment, housing, pensions, and medical care to its members, with workers being allocated to work units bureaucratically and wages being determined administratively. The urban wage structure was highly egalitarian. Movement from one work unit to another was extremely rare, and rural–urban migration was very limited. The slow evolutionary movement towards the formation of a labor market is analyzed and explained in Knight and Song (2005). The book is based mainly on a series of sample surveys, including the 1988, 1995, and 2002 CASS surveys referred to above.

The 1988 survey, conducted near the start of the urban reform process, reflected the administered and egalitarian labor system of the time. The Gini coefficient of urban wages rose continuously as reform has progressed. It rose from the low value of 23% in 1988 to 32% in 1995 (Knight & Song, 2007). Table 7 shows the percentage changes from one survey to the next. Whereas the median real wage rose by 6% per annum over this period, the corresponding percentages for the 10th and 90th percentiles were 1% and 8% respectively. Wage inequality continued to grow over the period 1995–2002: the Gini coefficient increased by an additional 5 percentage points, to 37%. Between those years the hourly real wage rose by 7% per annum at the median, 5% per annum at the 10th percentile, and 8% per annum at the 90th percentile. Given extreme disequilibrium in the urban labor market in 1988, this rise in wage inequality probably represented a move towards the market equilibrium rather than a movement of the market equilibrium.

Table 7.  Measures of central tendency and dispersion of the wage per worker at constant prices, urban China, 1988, 1995, and 2002
 Annual wage 1995 (1988 = 100)Hourly wage 2002 (1995 = 100)
  1. Source: Knight and Song (2007, table 9.1).

 10th percentile106137
 90th percentile175175
 Gini coefficient134114

The wage premium over primary school of a college degree was 9% in 1988, 39% in 1995, and 88% in 2002; for completion of secondary school, the corresponding figures were 4%, 17%, and 42%, respectively. A very similar story can be told about occupation-specific skills. We argue that some of the widening wage structure represents rewards for productivity and incentives for efficiency. Thus, for instance, the rising returns to education, the greater premia in skill-based occupations, and the more bowed inverse U-shaped wage-experience profile may well represent the stirring of market forces rewarding productivity.

On the other hand, there is evidence of greater wage discrimination (against women and in favor of Communist Party members) and also of considerable wage segmentation, which is difficult to justify in terms either of efficiency or of equity. The segmentation takes three main forms: segmentation by province, segmentation by type of ownership (with foreign firms paying most, then state enterprises, and local private firms paying least), and segmentation by profitability of enterprise. This last form we explain in terms of the danwei sharing profits with its urban employees (Knight & Li, 2005). Although many urban hukou workers are members of trade unions, the trade unions are subordinate to government and controlled by government: they have few of the rights and powers that trade unions in most countries possess. Nevertheless, it is possible that trade union representatives within an enterprise press for workers to share in rising profits, and that employers see informal profit sharing as a way to secure cooperation and maintain morale.

A decomposition of the growth of wage inequality between 1988 and 1995 showed that greater wage segmentation by province made the greatest contribution to that growth. An equivalent decomposition for the next 7 years found that segmentation by province, although still important, made a negative contribution to the growth of wage inequality, although another form of segmentation – by ownership type – did make a positive contribution (Knight & Song, 2007, table 9.5).

Labor mobility among urban hukou residents was still very low in 1999: 78% were still in their first job, 16% were only in their second, and the average duration of the first job (including incomplete tenure) was 21 years. (Knight & Yueh, 2004, p. 643). We conclude that China has a long way to go before it has a well-functioning labor market.

In summary, there is a good deal of support for the hypothesis that the rewards for productive characteristics rose during this period of growing labor market competition. However, the evidence also suggests that the labor market became more segmented – across firms, regions and probably also between urban and rural hukou workers. Both trends have contributed to rising inequality in the urban labor market.

5.3 Regional divergence

Two counteracting forces influence the degree of spatial income inequality. Processes of cumulative causation can create growth poles and so increase spatial inequality, whereas processes involving spillovers and spread effects can reduce it. The former is likely to hold sway in the initial stages of development but will eventually give way to the latter, as competitive advantages are eroded by rising costs.

Regional inequality in China has attracted a good deal of research, using conventional tests of absolute or conditional convergence or divergence. The results have been rather mixed, depending on the dataset, the dependent variable, the unit of observation, and the time period. There is much evidence of absolute divergence in the period up to 1995. Some scholars have found conditional convergence – a prediction of economic growth theory – but this is rather academic. Some of the variables being conditioned are endogenous, and economic welfare is anyway dependent on absolute trends.

Consider first the urban economy. There is evidence of powerful absolute divergence in urban income per capita and in urban wages among provinces and among cities over the period 1988–1995 (Knight et al., 2002). This is partly the result of regional comparative advantage, but it also reflects weak government regional equalization policy, imperfect capital markets, and continued restrictions on the mobility of labor. The coastal provinces gained not only from their advantageous location but also from initial preferential policies on foreign direct investment and exports and from the growth of tax revenues as their development proceeded. There is some indication that there have been elements of spatial convergence or at least stability since 1995. The same exercise is not available for the 1995–2002 period, but Gustafsson et al. (2007b, table 2.9) found that the percentage of inequality in urban China that is due to between-province inequality fell from 26% in 1995 to 19% in 2002. The main gain came from Eastern China, where this more developed regional economy was becoming more spatially integrated.

Turning to the rural economy, we start with a microcosm of rural society: the village. A small survey of seven villages in two counties in Hebei Province raised a fascinating question: how can villages that are so close together geographically be so far apart economically? Knight and Li (1997) developed an answer in terms of factor immobility and processes of cumulative causation. People cannot move freely among villages: each village is self-contained. Insiders have property rights, broadly interpreted, to village resources, and preclude others from sharing those rights. Institutional and informational problems segment capital markets, making the village itself the main source of funds for investment. Informational imperfections limit business and technical skill acquisition from outside. The village cannot rely much on the receipt of external public funds.

It is possible for a minority of villages to develop on the basis of fertile land that permits a prosperous agriculture, but the majority must rely on non-agricultural sources of development. One possibility is migration, but the opportunities are often limited. They can be expanded if the village can build up a social network of migrant contacts, which reduce informational and transaction costs. In that way migration from the village breeds more migration. The other possibility is the development of village industry. Here the important thing is to overcome the entrepreneurial and capital constraints. A good natural resource base helps. However, the development of village industry is very much a process that builds on itself through the generation of profits for reinvestment and of new skills via learning by doing. Funds for infrastructure investment depend on village taxable capacity, which in turn depends on the extent of collectively owned village industry. Path-dependent processes thus operate to propel some village economies forward and to hold some back.

The survey was conducted more than a decade ago, and the local economy may well have changed since then. The further growth of the more developed villages may have been retarded by shortages of labor and land and rising costs, and the less developed villages may have benefited from these positive feedbacks. With the rapid growth of urban China, the greater opportunities for rural–urban migration may also have assisted the poorer villages to kick-start the development process.

Evidence from the 1995 CASS survey shows the importance of non-farm employment for rural household incomes and for rural income inequality. Wages accounted for 21% of rural income inequality in 1988 (contributing 7 percentage points to the Gini) and for 44% (contributing 19 percentage points) in 1995 (Knight & Song, 2001, table 4.6). Wages rose from being 11% to being 26% of rural incomes over the 7 years. The contribution of wages to the rise in the Gini actually exceeded the total rise. It is clear that access to individual labor income was then the key to higher income for the household.

The Gini coefficient of rural incomes rose by only one percentage point between 1995 and 2002, or it fell by 4 percentage points (Table 2). Why the stagnation or reversal? Over that period the share of wages in rural income rose again, to 30%. However, the contribution of wages to rural inequality fell slightly, from 40% to 36% of the total (Khan & Riskin, 2007, table 3.3). Thus, wages became less disequalizing. The authors suggest that this is because the growth of wage employment reduced spatial, especially province, inequality of access to wage employment, and regard it as an important reason why the rural Gini stopped rising or started falling. It appears that spillover effects were becoming more important.

5.4 Rural–urban migration

Whereas the main absorber of the growing rural labor force in the 1980s was rural industry, since 1990 it has been rural–urban migration. The control of permanent migration through the hukou residence registration system means that much of this is temporary migration, with workers returning to their rural homesteads. It is very difficult to measure the number of rural migrants working in the urban areas but the 2000 population census suggests over 80 million.

This migration has differing effects on income inequality. The average income per capita of rural–urban migrant households is higher – nearly double – that of rural households, but about 35% lower than that of urban households (Gustafsson et al., 2007a, p. 45). Using a six-province urban survey of 1999, Appleton et al. (2004, table 1) found that the median wage of migrants was 54% of the median for (non-retrenched) urban hukou workers. However, a measure of the extent of labor market segmentation between urban and migrant workers requires standardization for productive characteristics. On that basis, Appleton et al. (2004, table 3) estimated that migrants received 75% of the urban wage. With the migrants being at an economic disadvantage and suffering official discrimination in the cities, the presence of migrant households raises urban inequality – provided, of course, that they are included in the urban household surveys. The 2002 CASS survey is the first national survey to permit estimation of urban inequality with migrants both excluded (the usual practice) and included. Khan and Riskin (2007, tables 3.6, 3.7, 3.10) estimate the urban Gini coefficient for household income per capita to be 32% when rural–urban migrant households are excluded and 34% when they are included; migrant households themselves have a Gini of 38%. These figures may understate the effect of migrants in raising urban income inequality because most of the rural–urban migrants – those not living in households – were not recorded in the survey.

The effect of remittances on rural incomes should reduce the urban–rural income disparity. Within rural areas, the effect on inequality depends on who migrates. The really poor may not have the education, or the funds, or the contacts, to migrate. However, such studies as have been made of this issue suggest that, in the Chinese case, migration has an equalizing effect on rural income distribution (for instance, Knight & Song, 1999, chapter 9).

The Lewis model (1954) provides a helpful theoretical framework for considering whether the increased rural–urban migration that accompanies economic development will eventually generate market forces that will raise labor incomes generally and, therefore, reduce income inequality. Lewis distinguished two stages in the development process: the labor-surplus stage, in which rural labor is abundant and available to the urban sector with an elastic supply curve, and the labor-scarcity stage, in which the rural labor supply is upward-sloping. When the turning point is reached, both the market-determined urban wage and the rural incomes, reflecting the rural supply price, begin to rise, and relatively unskilled labor incomes grow throughout the economy. If it is the case that prior to the Lewis turning point the labor market is segmented, with urban wages being above the level that would be set by market forces, when rural labor decreases relative to the other rural factors of production, labor incomes will be pulled up and the urban–rural income gap will begin to close.

Consider the application of the model to China. While China remains in the labor-surplus stage, we know that there is a considerable income gap between rural and urban people. However, migration has increase rapidly since 2000. The number of people in rural China began to fall in the mid-1990s. Cai et al. (2007), using the Ministry of Agriculture's household panel survey, estimate the number of rural workers in the urban labor market to be 78 million in 2000, when they accounted for 37% of the urban labor force, but no less than 132 million in 2006, accounting for 47% of the urban labor force. These authors conclude from this trend, and from reports of labor shortage and rising wages in the coastal areas which employ many migrants, that China is now close to the turning point. However, Meng and Bai (2007) provide evidence inconsistent with this conclusion. Because of the vast underemployment of labor in China's interior provinces, rising migrant wages may well be local and temporary on account of labor immobility rather than general and permanent (Knight, 2007). Migrant wages can rise either endogenously in response to labor shortages or exogenously in response to independently rising rural incomes. As Li Shi points out in his comment below, rural income per capita has recently risen on account of more supportive agricultural policies, and this will have raised the supply price of rural–urban migrants. Real income per capita of rural households as a whole rose by 36% between 2001 and 2005 (deducting income transfers makes no difference). Over the same 4-year period the hourly real wages of migrants in a five-city survey rose by 32%, and by only 22% in the case of those with no more than middle school education (Park et al., 2007).

The issue remains clouded by the lack of a reliable index of migrant wages, at both province and national levels. Nevertheless, the continued rapid development of the urban economy will at some time in the future unleash powerful market forces that will raise rural incomes, and labor incomes, generally, and thus reduce national income inequality.

5.5 Entrepreneurship, rent-seeking, and corruption

A rich entrepreneurial class has emerged remarkably rapidly in China – there were large supernormal profits to be earned. The combination of semi-marketized economy, weak legal system, and ill-defined or insecure property rights provides great scope for corruption, cronyism, rent-seeking, and appropriation of state assets. Corruption is a major concern of Chinese citizens as reported in public opinion surveys, especially where people are directly affected, such as in the provision of health services or the inadequately compensated loss of land. However, not all income derived from rent-seeking or corruption is detectable in the household surveys.

Both the NBS and CASS surveys may understate incomes at the top of the income distribution. One reason is the tendency for rich households to avoid inclusion in the NBS samples, either because of their higher opportunity cost or because they have more things to hide. The other is the tendency for rich households, if they are included, to underreport their incomes.

An ingenious method has been devised to measure the latter effect (Wang, 2007). The argument is that people have an incentive not to report “grey income” (i.e. illegal, corrupt, or dubious income), often reflecting the rents accruing to people with power over resource allocation in a semi-marketized economy. The hypothesis is that households with high legitimate income have greater opportunities to obtain gray income. To test this hypothesis, professional interviewers were employed to interview 2000 households in the normal way in several cities in 2005. The methodological trick was to get them to select only households whom they knew, either as relatives or friends, in the expectation that such people would be more honest in their answers to interviewers who were close to them.

Wang also had access to the corresponding income and expenditure information in the 2005 NBS urban household survey. The latter sample, ranked by income per capita, was divided into seven groups. The Engel coefficient (food consumption as a proportion of total consumption) and the mean income per capita of households in each group were calculated. It was assumed that the Engel coefficient could be measured without bias. The expected monotonic inverse relationship between the Engel coefficient and mean income per capita was found across the groups. Wang divided his own sample, ranked by income per capita, into seven groups so that each group had the same Engel coefficient as the corresponding group in the NBS sample. He then compared mean income per capita of the corresponding groups. The ratio of income per capita in the small survey to that in the NBS survey rose monotonically, from 1.1 in the lowest income group to 1.7 in the median group, to 2.0 in the second-highest, and to 4.9 in the highest. It is in the highest income group in particular that underreporting occurs. Income inequality among urban households is likely to be underestimated for this reason.

6. Was it Inevitable that Inequality Should Rise so Much?

Some of the rise in inequality was inevitable as the central planning system was dissolved and marketization occurred. However, some of the rise was due to the government's policy stance. As a reaction to the egalitarianism of the past, wherever there was a trade-off, efficiency considerations took priority over equity considerations. Government long-term strategy is well summed up in the words of a high official, Du Runsheng (1989, p. 192): “prosperity to few, then to many, then to all.”

I give three examples of how government chooses to trade equity for efficiency. One is regional policy. The most efficient location for most foreign direct investment is on the eastern seaboard. However, for a long time, foreign direct investment was required to locate there and prohibited from locating elsewhere. This gave the initially favored areas a headstart. Under central planning there was a substantial progressive provincial tax and transfer system that redistributed revenue from the richer to the poorer provinces. This redistribution was much weakened under economic reform, as each province relied increasingly on its own so-called extra-budgetary revenue and expenditure (Knight & Li, 1999). There was a reform of the fiscal system in 1994, intended to strengthen central government control of tax revenues. Nevertheless, public resources continued to be closely related to the level of local economic development. Richer areas have more revenues and are better able to provide public services and invest in local infrastructure. This worsens the spatial distribution of income. At the local level, villages, townships and counties are expected to pull themselves up by their own bootstraps. This degree of fiscal decentralization contributes to spatial inequality, and can produce poverty traps, for instance with regard to the quantity and quality of primary and secondary education (Knight et al., 2007). Spatial inequalities could be reduced if national and provincial governments were to adopt less decentralized and more equalizing fiscal arrangements. In this case, apparent efficiency is preferred to equity, but the playing field may not be sufficiently level to ensure efficiency.

A second example is the powerful urban bias which is apparent in much government policy, and which sustains and promotes the rural–urban divide. Urban bias exists in China, in two senses. First, government allocates fewer resources to the rural sector than it would if it were concerned only with improving economic efficiency, as determined by shadow prices. Second, rural-dwellers receive less priority than they would if the government social welfare function made no distinction between rural and urban residence per se, and placed a greater value on additional income, the poorer the person. Underlying urban bias is the greater potential threat to political stability and government survival that urban people constitute.

Being separately administered, the rural areas are at a great fiscal disadvantage: they are largely expected to find their own revenue. Agriculture has received little policy support, and rural industry – the great success of the early reform period – has grown relatively slowly in recent years as it has lost out to urban industry. This suggests that a low implicit weight is implicitly attached to the welfare of rural people in the government's social welfare. It is also partly the result of both fiscal and financial market arrangements. The high degree of fiscal decentralization within the rural economy makes it difficult for poor rural people and communities to better themselves. The deliberately repressed financial system, which rations credit and allocates it mainly to large state-controlled firms and does so inefficiently, imposes very tight credit rationing on rural people (Riedel et al., 2007, chapter 4). Rural investment is therefore probably sub-optimal even if judged solely by the criterion of efficiency. In this case, the urban bias policy may harm both efficiency and equity.

The third example is the large-scale redundancy program in the state sector that was introduced in the mid-1990s. This was forced on government by the declining profits and rising losses of state enterprises, as a result of growing competition, which threatened economic growth and state revenues. It did not herald the end of urban bias. Indeed, the urban workers who retained their jobs enjoyed wage increase of some 8% per annum throughout the period of the redundancy program. The redundant workers – more than 40 million of them – received little support and were thrown on the very tough residual labor market. There is a more general point, that the losers from economic reform have been neglected – as the social security previously provided by the danwei was tardily and inadequately replaced by more centralized systems. In this case, efficiency was achieved but without accompanying equity.

7. Does It Matter That Inequality Has Risen?

One can answer this question normatively or positively; here I take the latter approach. The Chinese economy is currently in a virtuous circle of high business and investor confidence, high investment, dynamic exports, and resultant rapid economic growth, which in turn justifies and sustains that confidence. That raises the question: could there be a shock to the economy which could destroy that virtuous circle? Here I limit myself to the threats that emanate from rising inequalities.

One threat is political instability. There are inevitably political tensions surrounding economic reform. They derive from the uneven benefits of economic growth, and from the pressures for a different distribution of those benefits. The regional inequalities, the rising urban–rural inequality and the policies of urban bias, the rising inequality of pay, the emergence of a wealthy entrepreneurial class, the public perception of rampant rent-seeking activities and of corruption, the losers from state enterprise reform, the urban tensions exacerbated by the flood of rural–urban migrants, the associated crime, the loss of social protection previously provided by the danwei, the rise in anomie among urban people confronted by new insecurities, and the weak system for managing industrial relations, are all potential sources of discontent. There is of course an authoritarian, one-party, repressive political system to keep the lid on discontent. But this system does not have the flexibility to deal with discontent and disorder except by means of rapid economic improvement or political repression. Things could boil over at some point.

It is fairly clear that government priorities are changing. When we began our income distribution research program in the late 1980s, government showed very little interest in it. Now my Chinese research colleagues are much in demand in government circles. The new leadership has begun to talk about “the harmonious society”, and some recent policy decisions like the Western development program and the abolition of school fees for rural compulsory education augur well. As against this, new interest groups are forming, which are likely to keep the economy on its inegalitarian course or weaken a policy response to discontent. For instance, even capitalists are now invited to join the Communist Party, and their influence on policy is probably growing.

8. Conclusion

In the early stages of economic reform, it was predictable that income inequality in China would grow rapidly, and indeed it did. The evidence for the last 10 years, since 1995, is more ambiguous. The CASS survey suggested that income inequality, already high, did not rise further, whereas the NBS data implied that inequality continued to grow. The difference may be due to the choice of income definition. Neither source provides a sure guide to the future of inequality in China.

Whether inequality rises or falls will depend on the balance of countervailing forces. It is clear that among the forces raising inequality in recent years are the growing importance of the rural–urban income disparity and the growth of urban wage inequality, itself to be explained both by greater rewards for human capital and by continuing labor market segmentation. Among the forces reducing inequality are spread-effects: the diffusion of non-farm wage employment in rural areas; some convergence in household income per capita among provinces, especially among those in the eastern region, and eventually China's general entry into the second stage of the Lewis model, when labour market forces will operate widely to reduce inequality from below. The relative importance of such economic relationships will determine future inequality, but they in turn will depend at least partly on evolving government objectives and policies.