1
Simon Appleton
School of Economics, University of Nottingham, NG7 2RD, U.K., email: simon.appleton@nottingham.ac.uk
Lina Song
School of Sociology and Social Policy, University of Nottingham, NG7 2RD, U.K., email: lina.song@nottingham.ac.uk
1
We are grateful to the CCK Foundation for the research grant and to Dr. Qingjie Xia for his invaluable research assistance.

1.
Introduction
Until the second half of the 1990s, policy-makers and researchers regarded urban poverty in China as either non-existent or a very marginal problem. Government antipoverty programs focussed on rural areas and, in particular, on selected poor counties.
As Khan (1998, p42) commented “
China’s official poverty reduction strategy is based on the assumption that poverty is a rural problem .” Within academia, few studies focussed on urban poverty, in contrast to the large literature on income distribution and inequality more generally. This neglect of urban poverty arose partly from using low poverty lines – such as a “dollar a day”. Due to the great urban-rural divide in
China (Knight and Song, 1999), a significant proportion of the rural population fell below these poverty lines but only 1% of the urban population were classified poor.
With urban poverty in China being defined so as to concern only a very small minority, it is scarcely surprising that the issue was marginalised by government and scholars alike. However, this attitude would be surprising to counter-parts in middle and high income countries, where poverty remains a relevant concept despite only very small proportions of citizens living under a dollar a day. Arguably, the main role of poverty lines is to focus attention on the living standards at the lower end of the income distribution. When set inappropriately low, they can have the opposite effect and help marginalise any discussion of the distributional effect of policy reform. This is what happened in urban China prior to the late 1990s.
Things began to change in the second half of the 1990s with concern over what was seen as a “new urban poverty” caused partly by a wave of rural-urban migration and partly by mass unemployment following a program of retrenchment in state owned

enterprises. These new forms of urban poverty differed from the old urban poverty which was often characterised as the “three withouts” – roughly corresponding to the disabled, the sick and the orphaned (Wong, 1998). By the turn of the century, opinion makers began to assert that urban poverty had risen during the 1990s, taking some of the shine of China’s exceptionally high rates of economic growth. For example,
The
Economist (2001, page 39) declared
“And in the cities, absolute poverty is increasing…” while the Chinese government magazine Liaowang (27 June 2002) also argued that urban poverty had increased. Underpinning such commentary was a concern that rising urban poverty would lead to political unrest, jeopardising the reforms that had enabled China’s rapid economic growth (see Wu, 2004). However, urban poverty in China has continued to be measured using low poverty lines such as
“a dollar a day” that, despite perceived adverse developments, still only categorised around 1% of the urban population as poor.
In this chapter, we use a range of higher poverty lines in order to consider more broadly how lower income urban households fared in the 1990s. Despite thus categorising a larger segment of the urban population as poor, we find that concern over adverse poverty trends in the 1990s appear misplaced. There has perhaps been an over-reaction to the previous neglect of urban poverty in China, with unwarranted pessimism about the living standards of poorer households. In particular, we challenge the assertion that urban poverty rose in absolute terms. While this claim makes dramatic headlines, it is does not appear to be supported by the evidence. In particular, we present new evidence on trends in urban poverty from 1988 to 2002 based on our analysis of the best available data-set on income distribution in China. The data come from the surveys in 1988, 1995, 1999 and 2002 conducted as part of the Chinese

Academy of Social Sciences (CASS) Household Income Project , which provides a much fuller accounting of household income than the biased measures used in official statistics. While undoubtedly there have been some losers from China’s urban reforms, the CASS surveys imply that this has not translated into a worsening of real income of those at the bottom of the income distribution. Indeed, while real incomes have grown less at the lower end of the distribution than higher up, they nonetheless have grown. As a consequence, whatever fixed point is taken as the poverty line, urban poverty has fallen. Although the existing literature on this question is mixed, we argue this result is consistent with official household statistics, despite the latter’s biased measurement of household incomes. Consequently, while we do not deny that the concept of the “new urban poverty” identifies new forms of relative poverty, the idea that these trends have led to absolute impoverishment appears to be something of a myth.
An important caveat to our argument is that the CASS surveys on which we base our estimates of urban poverty are based on the government’s official sampling frame.
While this has the advantage of making the surveys broadly representative of the country as a whole, it has the drawback of excluding the “floating population” of rural-urban migrants. Due to this data limitation, we – like all existing quantitative estimates of rates of urban poverty in China - confine ourselves to the study of poverty levels among Chinese residents with urban registration ( hukou ). Since ruralurban migrants are no doubt poorer as a group than residents with urban hukou , their exclusion will in some sense lead us to under-estimate urban poverty. Indeed, since rural-urban migration has increased in the 1990s, the exclusion of rural-urban migration will lead us to bias downward poverty rates in urban areas. However, this

limitation is perhaps less serious than it seems. Migrants have presumably moved to the cities to make themselves better off – indeed, some estimates put the return to migrant labour at ten times the marginal product of labour in agriculture (Song, 2000).
There is no real suggestion that migrants as a group have impoverished themselves by moving to the cities – if anything, the presumption is that migration has provided a means by which they can escape poverty (Park, Du and Wang, 2004). Contrary to the premise of the Harris-Todaro model of migration, what data we have on rural-urban migrants in China (for the year 1999) suggests that unemployment rates among them are negligible (Appleton et al., 2002). Hence even if we could get data on migrants, any rise in urban poverty due to their movement to the cities would surely be more than offset by a contrary decline in rural poverty. By contrast, the second half of the
1990s saw the emergence of mass unemployment among residents with urban hukous .
Employees in loss-making State Owned Enterprises found themselves laid-off and enduring long spells of unemployment. It is within this, formerly protected, group that a rise in absolute poverty is most commonly assumed to have occurred.
The structure of the chapter is as follows. Section 2 discusses some of the key issues – both the general debate about growth, poverty and inequality debate; and the specific case of China during its period of urban reforms. Conceptual and practical issues in the measurement of poverty are addressed in Section 3. The small existing literature on trends in urban poverty in China is reviewed in Section 4. Our own findings on household income regarding trends in growth, inequality and poverty are documented in Section 5. Section 6 summarises and concludes.

2.
Issues
2.1 The International Debate on Poverty Reduction and Growth
The study of the poverty trends during China’s rapid growth is of general interest because it informs a wider international debate. Approaches to poverty reduction differ in the relative emphasis they place on promoting economic growth and on redistributing existing resources in favour of the poor. One extreme would be to rely on economic growth – sometimes characterised as the “trickle down” approach, although this description implies a unjustified passivity in the role of the poor in bettering their own lot. At the other extreme, some argue that it is essential for the government to undertake large-scale redistribution through progressive taxation and welfare policies. How growth is actually distributed clearly has some bearing on this debate. If the poor did not benefit from growth alone, then redistribution would be necessary for poverty reduction. By contrast, if growth benefited the poor disproportionately, then the case for redistribution would be weakened.
The debate on poverty reduction and growth is particularly relevant to China, as its government has shifted its position during the reform period. In the era of planning, the Chinese state played a very strong redistributive role – for example, in urban areas, wage inequalities were compressed, unemployment avoided through implicit job guarantees and a variety of welfare benefits provided by work units. However, during the planning period, it could be some have characterised the Chinese government’s approach to poverty reduction as being closer to “trickle down” in emphasising growth over redistribution – in the words of Deng Xiao-Ping, “let some

get rich first”. In the interests of promoting economic efficiency and hence growth, enterprises have been given more freedom in letting worker remuneration reflect productivity, excess workers in stated owned enterprises have been made unemployed and many state transfers have been removed. As Khan (1998) argues, post-reform, the
Chinese government has tended to reject a “relief approach” to poverty reduction in favour of efforts to increase income generation. Indeed, Khan suggests that the reluctance of the Chinese government to adopt a redistributive approach to poverty reduction may partly be a backlash against the extreme egalitarianism of the planning period. Clearly, the efficacy of China’s current strategy of emphasising economic growth over redistribution depends partly on the extent to which growth has actually reduced poverty. However, despite the importance of the issue, theory and evidence provide suggest that a priori , we cannot be confident about the distributional effects of economic growth in a particular instance like that of urban China after 1988.
Early theorising about the relation over time between growth, inequality and poverty centred on Kuznets (1955) “inverse U” hypothesis. This postulated that inequality should rise with initial development, but then subsequently fall as development continued. The hypothesis was based on initially rather fragmentary cross-country evidence; Kuznets made the hypothesis after examining data on income distribution from only six countries at different levels of development and only at one point in time. While simple bivariate correlations using cross-country data appeared to corroborate Kuznets’ hypothesis, further work has shown that this hypothesis is not a valid empirical generalisation and specifically is not robust to controlling for a “Latin
America dummy”. Middle income countries are found disproportionately in Latin
America, where the distribution of income is particularly unequal – in part due to the

very uneven division of land. Without regional controls, this gives the appearance that being middle income leads to inequality, when in fact it is simply being a Latin
American country that is associated with higher inequality. Analysis of different country experiences in the post-war period generally implies that economic development does not have a systematic effect on inequality, if country-specific effects are controlled for (Fields, 1991). However, this generalisation masks a wide range of experiences, implying the desirability of monitoring the distribution of growth in countries of interest.
Kuznets’ provided some theoretical support for his hypothesis using a simple theoretical model. The model postulated that growth took the form of moving labour from a low return traditional sector to a high return modern sector. This paradigm has some relevance to inequality in China as a whole with the traditional sector perhaps being equated with agriculture and the modern sector with non-agricultural production. However, while relevant to China as a whole, it is not clear to what extent this characterisation is germane to inequality within urban areas
2
. An alternative use of Kuznets’ model might be to characterise the movement of urban labour out of lossmaking state owned enterprises into more economically viable companies, whether in the state or private sector.
To the extent that cross-country empirical evidence implies that, on average, there is no effect of development on inequality, it follows that one should expect growth to tend to raise the incomes of the poor on a one-for-one basis. Cross-country evidence appears to support this (Dollar and Kraay, 2000). Simple economic theories of growth

are also consistent with a favourable effect of growth on poverty. For example, within the Solow (1956) neoclassical growth model, growth in income per capita may arise in the long-term due to improvements in total factor productivity (TFP) or, in the short term, due to increased rates of accumulation of physical capital. TFP growth may be as a result of improved efficiency – perhaps due to marketisation, state reform or greater international competition – or due to improvements in technology. However, whether growth is due to improvements in TFP or to capital accumulation, the model predicts that wages will rise and, with them, the main income source of the poor. The
Solow model can also be augmented to allow human capital accumulation to play a role. In the Chinese case, the gradual improvement in the educational attainment of the urban population is likely to lead to increased returns to labour in general, although perhaps a fall in the return to education ceteris paribus .
However, while cross-country evidence implies that, on average, growth appears to reduce absolute poverty and have no impact on inequality, there is considerable variation in country experiences. Rapid economic growth in other East Asian economies since 1960 appears to have been progressively distributed and to have sharply reduced poverty. Conversely, there are examples of worsening income distribution during growth and indeed of the incomes of the poor falling
3
. China, by virtue of its large population and sustained high economic growth, is arguably the most important recent case for study of the relations between growth, inequality and poverty. Poverty trends in rural areas are, of course, important for a complete
2 It may throw some light on the impact of rural-urban migration in China, although the application is not direct. For example, Kuznets’ model assumed that labour moving out of the low return sector into the modern sector (i.e. migrants) is paid the same as those already in the high return sector.
3 The UK and the US in the 1980s are sometimes claimed to be instances of this. For developing country examples spanning shorter time periods, see the case of Brazil discussed by Datt and Ravallion
(1992) and the case of Uganda reported by Appleton and Ssewanyana, (2003).

treatment of this topic – but consideration of developments in urban areas is a necessary part of the analysis.
2.2 The distributional effects of China’s urban reforms
Since international evidence provide no firm predictions on the likely distributional effects of China’s economic growth after 1988, it is important to consider particular features of China’s experience that may be relevant. Perhaps the most important feature is that the period in question has seen a gradual transition from a predominantly planned economy to one that is more market led. Nee’s (1989) “market transition” hypothesis postulates that the move from a command economy to a competitive one will see a fall in returns to certain unproductive characteristics of workers – such as political connections – and a rise in returns to productive characteristics such as education. Conceivably, this may lead to a fall in wages of those with unproductive characteristics – e.g. the relatively unskilled – and ceteris paribus this may contribute to increased poverty. Whether this trend leads to a rise in absolute poverty, or merely relative poverty, will depend on the extent to which marketisation leads to general rises in wages, as well as changes in wage differentials.
However, the marketisation hypothesis has been disputed by those who argue that, far from leading to competitive outcomes, reform may at least initially lead to managerialism and the freedom of employers to vary wages according to a variety of factors, not just those affecting productivity. In such a situation, there may be more scope for rent-seeking behaviour, clientelism and other inequalities that cannot be justified on “efficiency” grounds. This may explain why urban wage data shows a rise in the premium for being a Communist Party member (Appleton et al., 2004).

Somewhat contrary to the predictions of the “market transition” hypothesis are those surrounding the impact of globalisation and increased international trade on China. A characteristic of China’s rapid economic growth is that it has been associated with an increase in exports and a general increase in the openness of the Chinese economy to international trade, most clearly signalled by China’s ascension of to the World Trade
Organisation. Wood (1995) uses a modified version of the Heckscher-Ohlin trade theory to make predictions about the distributional consequences of increased trade between rich countries and poor countries like China. He uses the theory to show that in poor countries, where unskilled labour is relatively abundant, one should expect a rise in the returns to unskilled labour and fall in the relative returns to skilled labour.
This trend may be enhanced by the increase in the supply of human capital noted earlier. Both developments should tend to make the distribution of wages more equal.
However, some there are countervailing trends– for example, the introduction of skillbiased technological change (e.g. computerisation).
Changes in the relative wages are only one way via which marketisation may affect poverty. A potentially more serious path is through unemployment. In the second half the 1990s, the urban reforms included a programme of mass redundancies in lossmaking State Own Enterprises. This led to a large rise in unemployment, which had previously been negligible. In a competitive labour market, one would expect such mass lay-offs to lead to falls in real wages and hence an increase in the demand for labour, ultimately reducing unemployment. However, real wages in China continued to rise strongly in China during the period of retrenchment and high levels of unemployment have prevailed, although they are now falling. It has often been argued

that this rise in urban unemployment is a key aspect of a “new urban poverty” in
China and may lead to widespread dissatisfaction, possibly triggering social and political instability. To the extent that the unemployed eventually find re-employment, there has also been concern that they will be re-employed on less favourable terms than other workers and hence that we will see the rise of a two-tier labour market.
Reform of the state owned sector is primarily viewed in terms of changes in ownership and incentives, and in responding to competitive pressures from lower cost producers. However, it may also mask a wider sectoral transformation of the Chinese economy as labour switches from primary production and traditional heavy industry to more services and higher tech production. The state enterprises that have shed the most labour tend to be in certain heavy industries, for example steel-making, while others producing manufactured goods for export to Western markets have thrived.
This transformation may be partly driven by the introduction of labour-saving technology in certain industries but also reflects changes in the pattern of demand as
China develops and is better able to penetrate world markets for consumer goods. In the short run, these changes in the fortunes of different industrial sectors will be mirrored in the fortunes of those working in these sectors. Urban poverty might be expected to rise among people working in contracting industries, as their wages and/or employment falls.
Another aspect of marketisation has been wholesale reform of state transfers system and what may loosely be termed the welfare system in China. There has been a trend towards the removal or, in some cases, monetisation of in-kind benefits, for example food coupons and housing subsidies. Compared to Western-style welfare systems, the

pre-reform Chinese welfare system was less focussed on poverty reduction and targeting transfers to the poor. However, even if the transfers previously provided by the state were fairly uniform across the income distribution, their removal would ceteris paribus be regressive.
3.
Measuring poverty
Conceptually, poverty exists when a person fails to attain a minimum level of well being or welfare. However, what constitutes well being in practice and what a minimum level of well being is are subjective issues that are widely debated; we discuss both in turn. We also address the issue of aggregation, although this is one area where there is more of a consensus.
3.1 Choice of welfare measure
In this chapter we use a money metric measure of welfare – specifically income per capita. This meshes with the traditional economic conception of poverty as deprivation in terms of basic material goods and services (food, housing, clothing, etc.). Income poverty is likely to be a major determinant of many material deprivations. Indeed, household income may be the single most comprehensive measure of material well being that does not rely on arbitrary weights 4 . It is typically positively correlated non-monetary dimensions of welfare, such as education and health. Nonetheless, it is clear that the match is likely to be far from perfect (Appleton

and Song, 1999). We do not dispute the multidimensionality of poverty and welfare as stressed, for example, by World Bank (2001). However, it not possible to do justice to all these dimensions aspects in a single chapter and attempting to do so ultimately risks such work being superficial. The potential importance of such other dimensions should be born in mind – for example, when considering the impact of the rise in unemployment. Studies of self-reported happiness suggest that unemployment reduces happiness more by more than would be predicted merely by the loss of income it induces (Oswald, 1997). This may be because unemployment reduces a person’s sense of self-worth and leads to social isolation, aspect of deprivation that are omitted by money-metric measures.
Within the class of money metric welfare measures, researchers face a choice between focussing on income or on consumption. There are arguments for favouring consumption over income – principally that it may better reflect long term income.
However, we focus on income for data reasons - the CASS surveys do not have the kind of detailed information on expenditures required to reliably estimate expenditures. Working with income rather than expenditure does bring some advantages; for example, the decomposition of income by source can give insights on explanations for changes in welfare. The choice of income rather than expenditure is likely to have consequences for estimates of poverty. Since income typically exceeds expenditure, comparing it with a given absolute poverty line will tend to lead to lower estimates of poverty. Some work using the National Bureau of Statistics (NBS) data on urban household incomes and expenditures finds that incomes have risen more
4 One could attempt to construct broader composite welfare indicators, but such exercises invariably involve rather ad hoc and controversial weighting, while arguing losing valuable information through the aggregation process.

than expenditures during the period, reflecting a rise in savings rates (Meng et al,
2004).
Although money metric measures focus primarily on private income (or consumption), it is possible to make some allowance for the public provision of goods and services. Here, the CASS surveys have an advantage over the standard official measures of household income. The surveys account for the value of various state transfers and subsidies that were particularly important in 1988. Since these transfers have been eroded, excluding them and focussing on private earnings would risk overestimating income growth and poverty reduction. The most important elements of these transfers in 1988 were food subsidies through the coupon system and housing subsidies.
Both income and expenditure are measured at the household level. Some adjustment must be made for household size and for simplicity we focus on income per capita. It may be desirable to allow for economies of scale of consumption and the differing needs of members of different ages. However, there is no consensus on the appropriate “equivalence scales” to be used in making such adjustments (Lanjouw and
Ravallion, 1995). We classify individuals as poor based on their household income per capita – as such we cannot explore intra-household inequalities that may be important (Haddad and Kanbur, 1990).
In this chapter, we follow the conventions in the measurement of income proposed by
Khan et al. (1993). One important aspect is the inclusion of an estimate of imputed income from the rent of owner occupied housing. This imputed rent is also a

significant part of income as measured from the CASS surveys, but is omitted from official NBS figures on household income and expenditure. Rents rose during the
1990s, implying that a bias from their omission in the NBS estimates that is in the opposite direction to the bias from NBSs omission of various subsidies.
To adjust for changes in prices, we use the official province specific urban CPI figures. These allow for regional variations in prices. However, their usage here is subject to a number of caveats. The weighting of various items in the CPI may not be the most appropriate for low-income households. In particular, they are likely to spend a higher proportion on food, whose relative price increased in the period. In addition, further research is required to check whether the CPI fully captures changes in the price of some publicly provided goods and services – notably, the rise in user charges for health care and increases in school fees. These are important caveats to the substantive conclusions of this chapter that must be borne in mind, although for brevity they will not constantly repeated in what follows.
3.2
Setting the poverty line
Although the setting of a poverty line is central to the measurement of poverty, it is problematic because it depends so heavily on value judgements. There is no clear threshold level of income at which it is clear that someone below it is deprived whereas someone above it is not. To this extent, the issue is subject to the

philosophical “paradox of the heap” 5
. Moreover, different people are likely to have different judgements over what constitutes a reasonable minimum standard of living.
Fortunately, controversy can be avoided to some extent by looking for poverty incidence dominance – that is to see, examining whether conclusions about poverty can be drawn that are robust to the setting of the poverty line. Nonetheless, sometimes no dominance relations exist and then the location of the poverty line affects the conclusion drawn. Moreover, policy-makers and the public tend to focus on headline statistics for poverty rates, making the setting of the poverty line used in any publicly reported statistics important in practice.
There is no official urban poverty line in China as yet, although separate poverty lines have been set for various cities in order to determine eligibility for benefits (Minimum
Living Support). Statistics on poverty in urban China have tended to use what seem to be low poverty lines, often based around a calorific anchor. The dominant method is to estimate a “Cost of Basic Needs”, whereby a food basket giving a required amount of calories – typically 2100 per person per day – is priced and then a mark-up added to allow for non-food requirements. This approach is widely used in developing countries and estimates often correspond to around a “dollar a day”, where the dollar is per person per day in 1985 Purchasing Power Parity adjusted dollars (Ravallion,
Datt and van der Walle, 1991). For this reason, researchers sometimes use the “dollar a day” poverty line in lieu of estimating their own Cost of Basic Needs line. Most recently, Ravallion and Chen (2004), in collaboration with statisticians in the NBS, used the Cost of Basic Needs approach to derive an urban poverty line of 1200 yuan
5 The paradox refers to the difficulty of defining how many grains of sand constitute a heap. One grain does not, millions does. However, since adding a single grain of sand to a pile of several makes an imperceptible difference, it cannot change some grains of sand into a heap.

(2002) prices. At the time of writing, it is said that the Chinese government will in the future adopt this as its official poverty line (Ravallion, private communication).
There is some controversy over how much of the urban population of China is estimated as poor under such calorie-based poverty line. Most estimates find only a very small proportion of the population to be poor. For example, Ravallion and Chen
(2004) estimate that less than 0.5% of the urban population fell under their poverty line in 2001. Earlier estimates using a “dollar a day” poverty line also tended to find that less than 1% of urban residents were living in poverty in the 1990s. The main exception to this is work by Khan (1998) who derived a calorie-based poverty line of
2291 yuan in 1995 prices. With this line, CASS survey data imply 8% of urban
Chinese were poor in 1988, rising to 8.8% in 1995. Khan (1998, p8) criticises the
“dismally low poverty threshold” more commonly used, saying “To use such a poverty threshold is to start with the presumption that there is no urban poverty.”
In this chapter, we do not attempt to derive a calorie-based poverty line and so cannot adjudicate in the dispute over its level. However, we have sympathy with Khan’s criticism of the very narrow definitions of poverty conventionally applied to urban
China. One pragmatic concern is that the proportion identified as poor may be too small for a survey of the kind used in this chapter to reliably analyse. Sampling problems and measurement errors may be particularly influential when determining results for the extreme tail of the distribution.
More fundamentally, using a low poverty line may only be appropriate for some purposes - for example, when considering the great rural-urban divide in China
(Knight and Song, 1999). For example, while Ravallion and Chen (2004) estimate that

only 0.5% of the urban population in 2001 fell below their poverty line, the proportion was 12.5% for the rural population. However, restricting oneself to such a narrow definition of poverty means that what one concludes about the impact of growth or policy reform on urban poverty is of limited interest – pertaining as it would, in 2001, to less than 1% of the urban population. For example, in our CASS survey data, it would lead to no households surveyed in Beijing being classified as poor. Arguably, rising living standards in urban China have made the use of a calorific anchor inappropriate when setting the poverty line, much as it is inapplicable to OECD countries. Researchers studying poverty in industrialised countries would not dismiss the topic merely because everyone in such countries could afford to buy sufficient calories.
Consequently, we adopt a broader conception of poverty and do not rely on a calorie- based poverty line. We would argue that in urban China being unable to afford sufficient calories is not very widespread and is not the most useful approach to conceiving urban poverty. Instead, we view setting a poverty line as one way of focussing attention on the more disadvantaged people in urban areas. We examine dominance relations between poverty incidence curves, implicitly using a continuum of poverty lines. For our numerical estimates, we arbitrarily set the poverty line at $3, identifying 36% as poor in our data for 1988. The point to emphasise here is that the importance of our analysis is not so much the level of poverty at any one time when using this line, but comparisons of poverty over time and between groups (Ravallion,
1992). Essentially the poverty line is not intrinsically of particular interest, but provides a useful benchmark for analysing the living standards of the less advantaged.
Nonetheless, we also explore the implications of setting the line at $2 and use

dominance analysis to establish if our conclusions on poverty trends are robust to the use of different poverty lines.
A crucial issue in setting the poverty line is whether to hold it fixed over time and space – and hence in some sense absolute – or whether to allow it vary with the general living standard of the populations being studied – and hence relative.
Although we adopt rather arbitrary poverty lines, we hold them fixed over time and do not allow them to rise with overall economic growth. Taking such an absolute approach is essential if the figures on poverty are to reflect the living standards of the poor. For example, if we take a relative conception of poverty then it can appear that poverty rises despite the poor getting better off (so long as the non-poor get better off even faster). Holding the poverty line fixed – and hence making it absolute – can provide a good measure of changes in material deprivation, whereas a relative poverty may well be wholly misleading. We also do not allow for regional variations in the poverty line. Since we control for regional differences in prices, it is not clear that regional poverty lines are necessary (see Appleton, 1993, for a discussion on whether to use regional or national poverty lines). Nonetheless, it should be noted that, in
Sen’s (1983) terms, there is an absolute dimension to relative deprivation. If people have less material goods than others, they may be socially excluded and as a consequence experience real suffering. Indeed relative income, rather than absolute income, may be important in determining people’s satisfaction and hence in affecting socio-political instability. Given this interest in relative poverty, we also estimate the extent of poverty using a relative poverty line – specifically, half of median income in the year of the survey.

3.3
Choice of poverty indices
The final step in operationalising a measure of poverty is to adopt a specific poverty index that allows us to aggregate over the population. Here, we use the Foster-Greer-
Thorbecke (1984) P-alpha poverty indicators. This is a class of measures that includes the poverty headcount (the percentage of the population that is poor). However it also includes measures of the depth of poverty – how far below the poverty line, the poor lie – and of the severity of poverty, indices that are sensitive to distribution among the poor. The advantage of this class of measures is partly that it satisfies various desirable axioms for a poverty indicator. In addition, the indicators are additively decomposable, which is useful for seeing the contribution of various mutually exclusive groups (e.g. the employed and the unemployed) to overall poverty.
3.4. Data
This chapter will use urban Household Income Surveys conducted by the Economics
Institute, CASS, in1988, 1995, 1999 and 2002. The surveys are based on sub-samples from the main nationally representative household survey programme conducted by the National Bureau of Statistics. The samples are reasonably large: for example,
9,000 urban households from 10 provinces in 1988 and 7,000 urban households from
11 provinces in 1995 (the additional province being Sichuan). As previously noted, a serious limitation of the surveys is that they cover only residents with urban hukou, and so exclude rural-urban migrants. The excluded “floating population” was

estimated to constitute about 20% of all people actually in urban areas of China in
1995 and since then the proportion is likely to have increased.
The CASS Household Income Surveys were designed by a collaborative research team composed of Chinese and international scholars in order to analyse income distribution issues within China (results from the early surveys are reported in books by Griffin and Zhao, 1993; and by Riskin, Zhao and Shi, 2001). A key strength of the surveys is that they provide a more comprehensive and accurate assessment of household income than official NBS data (see Khan et al, 1993). Moreover, the surveys allow more in-depth and purpose-designed questions to be asked, allowing richer analysis. However, the surveys are smaller scale than the full NBS household survey effort and inevitably have a narrower geographic coverage. Nonetheless, the surveys are designed to be representative of urban areas in China as a whole.
4.
The literature on trends in urban poverty
There are relatively few studies of urban poverty in China and these have tended to present seemingly contradictory conclusions on recent trends. Some studies argue that poverty has increased in the 1990s; others that it has shown no trend and others that it has fallen. In this section, we review these studies and attempt to adjudicate between them. The task of adjudication is made somewhat easier by the fact that all rely on one of two main sources of data – either the official NBS household survey results or the
CASS Household Income Project surveys used in this chapter. We have argued above that the CASS data is preferable due to its fuller accounting of income sources,

however, it is not clear that a difference in data accounts for the conflicting results on poverty trends.
The result that poverty rose was first obtained by comparing the CASS surveys for
1988 and 1995 (Khan, 1998; Khan and Riskin, 2001). This work took the data on household income per capita from these surveys and compared them with a poverty line based on an estimate of the cost of basic needs. Anchoring the line on a requirement of 2150 calories per day, they estimated that 8% of urban residents were poor, rising to 8.8% in 1995. Taking a lower poverty line, anchored on 2100 calories per day, the rise was sharper – going from 2.7% to 4.0%. In some ways these results are the most comparable to those we present later in this chapter, as we take the same
CASS data, but add in surveys for 1999 and 2002. We do not dispute the analysis of the 1988 and 1995 survey, but show that the rise in poverty in that period was not sustained and indeed was clearly outweighed by the fall in poverty from 1995 to 2002.
It seems likely that the CASS surveys yield less favourable estimates of poverty trends that the NSB figures. For example, Khan (1996) used tabulated NBS data to estimate that poverty fell from 7.42% in 1989 to 0.84% in 1994. The discrepancy seems likely to be explained by the fact that only income as measured in the CASS surveys included various subsidies that were gradually withdrawn during the period.
However, apparent support for Khan’s finding of rising urban poverty in the 1990s is provided by the recent study by Meng, Gregory and Wang (2004) using NBS data. It also adopts a cost of basic needs approach to setting poverty lines, but compares it to expenditure per capita (results with income per capita are discussed to show robustness). A key feature of the study is that it estimates a different poverty line for

each year. Unlike most other studies, poverty lines are also allowed to differ by region. The authors do not provide numerical estimates of poverty rates, but their text and graphs clearly show that they estimate a higher poverty rate in 1999 than in 1986.
Some studies using official tabulations from NBS data have concluded that there has been no clear trend in urban poverty in the 1990s if defined according to a “dollar a day” poverty line. Fang et al. (2002) use a subset of the NBS data – one representative city from each province. They compare household expenditures per capita with poverty lines based on the “dollar a day” standard widely used for international comparisons. Using the dollar a day poverty line, they find urban poverty rates to be the same (2%) in 1992 and 1998. Taking a higher poverty line of $1.5 a day, they report clear evidence of a fall in poverty from 14% to 9%. Chen and Wang (2002) perform a similar exercise on the full NBS data-set but are dependent only on official tabulations (grouped data). They use the income rather than expenditure data as their welfare measures. In their analysis poverty appears to fluctuate over the 1990s if one takes a “$1 a day” poverty line. Although their headcount figure is 0.5% in 1999 compared to 1% in 1990, it is hard to conclude that poverty fell during the 1990s as in
1998 it stood at 1%. As with Fang et al (2002), using a higher poverty line would imply a clear fall in poverty. For example, with a $1.5 a day line, poverty falls from
8.6% in 1990 2.2% in 1999; with a $2 a day line, the fall is from 20.7% to 6.8%.
More positive conclusions are made in other studies of NBS income data. In particular, Wang, Shi and Zheng (2001) use interpolations from the officially tabulated grouped income and look for generalised Lorenz dominance from 1981 to
1999. They find that each year, with the exception of 1988, the generalised Lorenz

curve for real income per capita dominated that for the previous year. This implies that social welfare rose year on year and hence poverty fell, regardless of what level the poverty line was set – so long as the line was constant in real terms over time.
Ravallion and Chen (2004) also work with the tabulated NBS data and devise a poverty line for urban China based on the cost of basic needs in collaboration with government statisticians. Comparing household real incomes with the line they derive, they estimate urban poverty rates to have been 2.07 in 1988, falling to 0.85% in 1995,
0.57% in 1999 and 0.54% in 2002.
How can we adjudicate between these seemingly conflicting results? Arguably, the work on the CASS data is the most authoritative – providing the best fullest measure of income. To the extent that the period 1988 to 1995 saw a large reduction in government transfers, not adjusting for this is likely to lead to too optimistic an assessment of poverty trends. However, to date, analysis of the CASS data has been limited to 1995 and anyway, studies of the official NBS data reach conflicting results.
The conclusion by Meng et al. (2004) that poverty rose over a longer period, based on official NBS data is questionable given that studies re-estimation of calorie-based poverty lines for each year of data. In a period of economic growth such as urban
China has enjoyed, people are likely to consume more expensive food, raising the cost of calories. Thus the food poverty lines used by Meng et al. will be rising over time.
Moreover, non-food requirements are a mark-up based on the non-food share of the poor. By Engels’ law, rising income leads to a rising non-food share and thus increases in estimated non-food requirements. Hence the non-food component of the poverty lines used by Meng et al. is likely to be rising. Allowing poverty lines to rise

in this way may give a misleading impression of rising absolute poverty when in reality the real incomes of poorer urban residents may have been increasing.
Among the other studies, the tendency to focus on low poverty lines that estimate only a small proportion of the urban population to be poor is troubling. It is likely that the surveys are less reliable in obtaining estimates for the very poorest – there may be particular problems sampling them and results may be very influenced by low outliers. The difficulties are likely to be particularly acute given that several studies use only official tabulated data with rather crude groupings. This requires interpolation that may be particularly difficult when dealing with only the extreme low end of the distribution. When using low poverty lines – a dollar a day or thereabouts – it seems hard to reach a confident conclusion about trends in urban poverty from the NBS data. However, with somewhat higher poverty lines that are fixed over time, it seems clear that there has been an improvement in the living standards of low- income households during the 1990s.
5. New Evidence on trends in poverty, inequality and growth
5.1 Household income growth and inequality
Table 1 provides some basic information on household real income per capita in the surveys by decile. It can be seen that real incomes have risen substantially over the fourteen-year interval covered by the surveys. On average, urban Chinese households have around twice as much income per capita in 2002 as they did in 1988. The table

also provides a breakdown of income by source, using the Khan et al. (1993) definition of income. There is a marked change in the structure of income between
1988 and subsequent years. In particular, subsidies and income in kind constitute a much smaller part of total household income after 1988. Ration coupons were abolished and housing subsidies shrank from 18% of total income in 1988 to a mere
2.8% in 2002. This is important, as some studies of income inequality in China focus more narrowly on cash wage earnings. Neglecting to account for subsidies and inkind, which have been largely withdrawn after 1988, will tend to overstate the rise in income during the period. In our data, the share of cash earnings by working household members has rose from 43% of all household income in 1988 to 60% in
2002.
The more inclusive measure of income in the Household Income Project (HIP) probably explains the discrepancy between the growth estimates from that data compared to those from the larger household surveys conducted by the NBS. NBS data imply higher growth during the period – 6.8% per annum compared to 5.1%
(Table 2 refers). HIP data record slower growth in the period 1988 to 1995, as subsidies were withdrawn, and also 1999 to 2002, as housing items contributed a smaller share to total income. By contrast, HIP data imply higher growth between
1995 and 2002 as housing rental values rose rapidly. In passing, it is interesting to note that China’s real GDP per capita (nationally, not urban-only) is estimated to have grown by 7.4% between 1988 and 2002. The HIP data thus imply substantially less income growth than official data, although a sustained growth rate of 5% per capita is still very high by international standards.

The focus of this chapter is not on the average level of growth, but how growth has varied across the distribution of income and hence the impact on poverty and inequality. Table 3 reports income per capita at each decile and Figure 1 plots the implied annualised growth rates. The HIP data show that income growth between
1988 and 2002 is greater, the higher up the income distribution one goes. We have already noted that average incomes virtually doubled in the period. However, for the lowest decile, the 10 th percentile, real income per capita only increased by a half (49% higher). For the highest decile, the 90th percentile, incomes increased by 130%. As a consequence, growth rates for the highest decile averaged 6.0% per annum, more than twice the 2.8% growth experienced by the poorest decile.
This earlier interval of growth is largely what accounts for the unequal pattern of growth over the full period. Between 1988 and 1995, incomes of the poor grew slowly: the poorest decile saw only slow growth in income in this early period (0.8% per annum). By contrast, the top decile, the 90% percentile, enjoyed very fast growth of 6.1% per annum. In subsequent intervals, the pattern of growth across the deciles is much flatter and less marked. It is true that the growth is slower at the poorest three deciles than at the median – but the differences are more muted - particularly in the latest episode, 1999 to 2000. Income growth for the most affluent decile is also below the median in these intervals.
The fact that income grew less for poorer deciles than for more affluent ones implies an increase in inequality. This is demonstrated in Table 3 that presents a variety of different inequality indices. By any of the conventional indices, inequality rises substantially from 1988 to 2002. For example, the Gini coefficient rises from 0.24 to

0.33. However, it is noticeable that most of the rise occurs between 1988 and 1995 – for example, the Gini coefficient is virtually unchanged between 1995 and 2002.
Looking at the numbers in more detail, there is a modest rise in inequality between
1995 and 1999. This is more or less offset by a modest fall in inequality between 1999 and 2002. The slight fall in inequality between the last two surveys is perhaps surprising given the evidence in Figure 1 that the poorer deciles experienced growth below the median. However this is outweighed by the fact that the most affluent also enjoyed below average growth (in this respect, it should be noted that the information on the 90 th
percentile tells us nothing about the richest 10% of households who tend to take a large share of total income).
5.2 Poverty trends
The fact that incomes grew across the deciles implies that poverty fell, so long as a constant, reasonably broad poverty line is used. Figure 2 provides figures for the percentage of urban residents who are poor for a continuum of poverty lines. The graph shows a dramatic fall in poverty during the full period from 1988 to 2002. This is true whatever poverty line is chosen, provided the poverty line is held constant in real terms (i.e. measures absolute, rather than relative, poverty). The “dollar a day” poverty line corresponds to a value of 1212 yuan per year in 2002 prices, or the point
“12” on the graph. As can be seen, less than 1% of people received incomes below this point, so that poverty line is of little use in evaluating changes in the standard of living of the less advantaged Chinese residents. Instead, we focus on two rather arbitrary poverty lines – a broad poverty line of $3 per day and a narrower one of $2.

With the $3 poverty line, the proportion of urban people who are poor falls from
36.4% in 1988 to 8.5% in 2002 (Table 5 refers). With the $2 poverty line, the prevalence of poverty falls from 7.3% to 2.1%. However, Figure 2 shows that the conclusion that absolute poverty has fallen is robust to the choice of poverty line because the poverty incidence curve for 1988 dominates that for 2002. Whatever constant poverty line we take, the incidence of poverty is higher in 1988 than in 2002.
Furthermore, one corollary of the “first order” dominance revealed in Figure 2 is that poverty must also be lower in 2002 than in 1988 on any P-alpha poverty index, regardless of what value of alpha we choose. Thus we can say that the poverty gap
(P1) and the squared poverty gap (P2) are lower in 2002 than 1988, irrespective of the poverty line chosen.
Poverty also falls between each of the surveys for most poverty lines. The most noticeable case where the poverty incidence curves in Figure 2 cross is when comparing poverty in 1988 and 1995. Here, for low poverty lines that identify less than 6% of people as poor in 1988, we can see that poverty is higher in 1995 than in
1988. This implies that living standards worsened for the poorest 5% of the population between 1988 and 1995. This helps to understand the finding in Table 5 that, using the $2 a day poverty line, the poverty gap, P1, and the squared poverty gap,
P2, are estimated to rise between 1988 and 1995.
Table 5 also reports poverty indices adopting a relative approach to poverty.
Specifically it presents poverty indicators for when the poverty line is defined to be one half of median income per capita in the survey of that year (i.e. allowing the poverty line to rise with growth). Since the incomes of poorer urban residents have

tended to grow less than average incomes, relative poverty has risen. Under this approach, relative poverty rises from 3.8% in 1988 to 12.8% in 2002. Relative poverty rises in all years although the increase is more pronounced between 1988 and 1995
(when relative poverty reaches 9.3%) and more modest between 1999 and 2002
(going from 11.8% to 12.8%).
5.3 Decomposition of poverty changes into growth and distribution components
The problem with focussing on relative poverty is that by construction it does not allow changes in average income to impact on poverty – relative poverty can only change if the distribution of incomes changes. However, it is growth alone rather than redistribution that has raised the living standards of the poor in this period. Since inequality has risen, the distributional changes have been unfavourable to the poor.
Table 6 decomposes the change in poverty into growth and redistribution components following Datt and Ravallion (1992). For brevity, we decompose the headcount (P0) index only – decompositions for the P1 and P2 are not reported in the tables but give qualitatively similar results. The strong impact of growth in reducing poverty is evident from the results. For example, using the three dollar a day poverty line, we find that if the poor had enjoyed the same rate of income growth between 1988 and
2002 as the means of our samples, then the percentage of urban Chinese who are poor would have fallen by 34.9 percentage points. Since only 36.4% lived under $3 a day in 1988, such growth would have implied the virtual elimination of poverty as so defined. In reality, poverty fell by 27.8 percentage points – still very impressive, but

short of what would have happened had there been no change in the distribution of income.
Table 6 shows that the distributional changes in income in the period have generally been unfavourable – as should be expected from the rise in inequality noted in Table
4. More revealingly, the impact on poverty of adverse distributional changes is estimated to have been substantial. For example, Table 6 implies that had there been no growth in mean income between 1988 and 2002, the worsening of the distribution of income would have increased the headcount by 8 percentage points (a rise in poverty of over a fifth). The only exception to the unfavourable distributional changes is the period 1999-2002 when the income distribution improves slightly and so would have implied a reduction in poverty even without growth.
The decompositions in Table 6 do not provide a very clear picture of which component – growth or redistribution was more important in determining poverty trends. This is perhaps surprising given the strong growth and fall in poverty despite adverse distributional changes. It is true that if one uses the $3 a day poverty line, the growth component of the decomposition is far larger than the redistributional component. However, the reverse is true when one uses the $2 a day poverty line.
This can be reconciled with the actual fall in poverty by noting that the residual component of the decomposition is often large and negative. Unfortunately since the residual component is merely a balancing item, it has no clear interpretation. Perhaps the main implication of Table 6 is that both growth and redistribution have been sufficient large to have had significant poverty impacts on urban China in the period, although fortunately favourable developments have predominated.

As might be expected, the redistributional component of the decomposition of poverty changes is most marked when considering the interval between the 1988 and 1995 surveys. For the narrower definition of poverty ($2 a day poverty line), the distributional component of the poverty changes is almost fully twice the size of the growth component. This implies that, for the very poor, adverse changes in the distribution of income outweighed the beneficial effects of general economic growth.
5.4 The paradox of rising unemployment and falling poverty
Perhaps the most surprising aspect of our results is that poverty in absolute terms has fallen at the same time as mass unemployment has emerged. Table 7 provides some insight into this paradox, presenting poverty statistics for 1995 and 2002 for households classified according to the economic activity of the household head. Of particular interest is the comparison between households headed by employed workers and those headed by the unemployed. However, the table also reports on those headed by the retired and by those who do not participate in the labour market for other reasons (e.g. attending domestic duties). Given these mutually exclusive groups, it is possible to decompose the overall change in the proportion living in poverty into the effects of changes in poverty within the groups and changes in the size of each group (Ravallion and Huppi, 1991). Specifically, if P t is the total poverty indicator at time t and P it
the corresponding indicator for those belonging to a group i, then:

P
T
-P
0
=

(P
+

(n
+

(P iT iT iT
-P i0
)n i0
-n i0
)P i0
-P i0
)(n iT
– n i0
) intra-group effects inter-group effects interaction effects where n it
is the proportion of the population in group i at time t. The interaction effects would be positive if people moved into groups where poverty was falling.
The impact of the program of lay-offs in the state sector in the second half of the
1990s is shown in the population shares of the various groups in Table 7. In 1995, only 0.4% of individuals in the sample lived in households headed by an unemployed worker. In 2002, this percentage had risen to 6.2%. Perhaps even more revealingly, the proportion living in households with employed heads fell from 80% to 71%.
Although one might expect some increase in the proportion living in households with retired heads due to an ageing of the population, our figures are consistent with some of the retrenchment in China having taken the form of early retirement rather than unemployment per se.
Other things being equal, the emergence of mass unemployment would be expected to increase poverty. This is born out by the contribution of the population shifts shown in Table 7. For example, Table 7a uses the “two dollar a day” line. The results imply that the poverty headcount would have increased by 1.3 points, from its 1995 level of
7%. However, this is more than offset by falls in poverty rates within groups. For example, the fall in the headcount among those living in households with employed heads would imply a 4.1 point drop in the poverty headcount. This alone would account for four-fifths of the observed fall in poverty. Moreover, the interaction effects also imply falls in poverty because the groups that have grown in size – those headed by the unemployed and the retired – have also experienced the fastest

reduction in poverty. The results are qualitatively similar when using the three dollar a day line, as in Table 7b.
Perhaps the most important fact why the rise in unemployment is not as disastrous as might be thought because only a minority of households headed by the unemployed are poor. For example, using $2 a day as the poverty line, 7% of people in households headed by the unemployed are poor in 2002. This is only a small minority, even though substantially above the 2% poverty headcount for the total urban population. If
$3 a day was used as the poverty, 22% of those in households headed by the unemployed would be poor. These figures are all the more remarkable because our poverty rates are measured by income, rather than consumption. Clearly, households with unemployed heads are finding sources of income other than their heads earnings to support themselves. This income is partly earnings from the spouse of the head (or other family members). In this respect, the high rates of female employment in urban
China should be acknowledged.

6.
Conclusion
China’s high economic growth since reform is perhaps the most important economic development in the world in the last quarter of a century. Although much of its significance has been in how it has transformed the lives of people living in rural areas, it is important also to examine its implications for urban areas – areas which accounted for 39% of China’s population in 2002. There has been a growing unease that this growth has been unequally shared and has led to the rise of a “new urban poverty”. Under this perception, economic efficiency and growth may have been promoted by urban reforms involving a reduction in subsidies for urban households and retrenchment of excess employees in State Owned Enterprises. However, these same reforms may have increased urban poverty, by reducing transfers to low income households and inducing mass unemployment.
In this chapter, we have focussed on the real incomes of urban residents at the lower end of the income distribution in urban areas. Using CASS surveys which include state subsidies and transfers in their measurement of household income, we have shown that living standards rose across the distribution of income from 1988 to 2002.
We do find evidence that the withdrawal of subsidies between 1988 and 1995 lowered the real income of the poorest in urban areas. However, this was subsequently outweighed by growth in other sources of income. Perhaps most surprisingly, we find that – despite high unemployment after 1995 – absolute poverty continued to fall, irrespective of where the poverty line was set. This implies that the concern that absolute poverty has risen during urban reform is misplaced.

This is not to imply that there have been no losers from economic reform.
Unemployment obviously adversely affects those subject to it. However, most households whose heads have suffered unemployment have not been driven into poverty, even when the poverty line is defined well in excess of the low “dollar a day” thresholds commonly applied to urban China. Here, a limitation of our money metric approach to poverty and well-being should be recalled. We may be understating the costs of unemployment to the extent that it causes suffering not captured purely by the loss of income – for example, social isolation and a loss of self-worth. Nor do our findings deny that relative poverty has risen. Although lower income urban households have experienced improvements in living standards, the rates of growth have been below those enjoyed by more affluent households. Since relative poverty and unemployment are both likely to be of concern to people – and hence may be a source for potential social unrest - our results on absolute poverty should not be taken as an argument for complacency when reviewing China’s economic progress. Nor have we touched upon the rural-urban migrants, whose living standards are much lower than those of urban residents. Nonetheless, it is important to challenge the idea that absolute poverty rose during China’s reforms before it becomes a new urban legend.

References
Appleton, Simon (2003), “Regional or national poverty lines? The case of Uganda in the
1990s”,
Journal of African Economies 12(4):568-624.
Appleton, Simon and Sarah Ssewanyana (2003) “Poverty estimates from the Uganda
National Household Survey II, 2002/2003”, mimeo, University of Nottingham, UK.
Appleton, Simon, John Knight, Lina Song and Qingjie Xia (2002) “Labour
Retrenchment in China: Determinants and Consequences”,
China Economic Review,
13(2-3): 252-275.
Appleton, Simon and Lina Song (1999), “Income and human development at the household level”, Background paper for the
World Development Report 2000/2001 ,
World Bank: Washington DC.
Chen, S. and Y. Wang (2001)
China’s growth and poverty reduction: recent trends between 1990 and 1999
WBUPDS seminar on “Strengthening poverty data collection and analysis” Manila: Philippines.
Datt, G. and M. Ravallion (1992) "Growth and redistribution components of changes in poverty measures: a decomposition with application to Brazil and India in the
1980s" Journal of Development Economics 38:275-295.
Dollar, David and Aart Kray (2000) "Growth is good for the Poor", World Bank
Working Paper, available to download from www.worldbank.org/research .
Fang, Cheng, Xiaobo Zhang and Shenggen Fan (2002), “Emergence of urban poverty and inequality in China: evidence from household survey”
China Economic Review
12: 430-443.
Fields, Gary (2001) Distribution and Development , Cambridge, MA: Russell Sage
Publishers.
Foster, J., J. Greer and E. Thorbecke (1984), "A Class of Decomposable Poverty
Indices", Econometrica , Vol. 52, pp. 761-765.
Haddad, L. and R. Kanbur (1990), "How Serious is the Neglect of Intra-household
Inequality?", Economic Journal , Vol. 100, pp. 866-881.
Khan, Azizur (1996) The impact of recent macroeconomic and sectoral changes on the poor and women in China New Delhi: ILO.
---- (1998) “Poverty in China in the era of globalization”, Issues in Development
Discussion Paper 22 International Labour Organisation: Geneva.
Khan, A.R., K.Griffin, C. Riskin and Zhao Renwei (1993) “Household income and its definition in China”, chapter 1 of Griffin, Keith and Zhao Renwei (eds.),
The
Distribution of Income in China Macmillan: London.

Khan, A. and C. Riskin (2001) Inequality and poverty in China in the age of globalization Oxford University Press: New York.
Knight, John and Lina Song (1999) The rural-urban divide: economic disparities and interactions in China , Oxford University Press: Oxford.
Kuznets, Simon (1955), “Economic growth and income inequality”
American
Economic Review 45(1): 1-28.
Lanjouw, P. and M. Ravallion, (1995) "Poverty and household size", Economic
Journal , 105 (433): 1415-1434.
Meng, Xin, Robert Gregory and Youjan Wang (2004) “Poverty, inequality and growth in urban China, 1986-2000”, paper presented at a conference on “Poverty,
Inequality, Labour Market and Welfare Reform in China”, 25-27 August, Australian
National University: Canberra.
Nee, V. (1989) “A theory of market transition: from redistribution to markets in state socialism”, American Sociological Review 54 (5): 663-681.
Oswald, A. (1997) “Happiness and Economic Performance",
Economic Journal , 107:
1815-1831.
Park, Albert, Yang Du and Sangui Wang(2004), “Is migration helping China’s poor?”, paper presented at a conference on “Poverty, Inequality, Labour Market and
Welfare Reform in China”, 25-27 August, Australian National University: Canberra.
Ravallion, M. (1992) “Poverty comparisons : a guide to concepts and methods”,
Living Standards Measurement Paper No. 88, World Bank: Washington DC.
Ravallion, M. and S. Chen (2004), “China’s (Uneven) Progress against Poverty”, paper presented at a conference on “Poverty, Inequality, Labour Market and Welfare
Reform in China”, 25-27 August, Australian National University: Canberra.
Ravallion, M., G. Datt and D. van der Walle (1991) “Quantifying absolute poverty in the developing world”
Review of Income and Wealth 37(4): 345-361.
Ravallion, Martin and Monica Huppi (1991), "Measuring changes in poverty: a methodological case study of Indonesia during an adjustment period", World Bank
Economic Review 5(1): 57-82.
Riskin, Carl, Zhao Renwei, and Li Shi (eds.) (2001) China’s Retreat from Equality:
Income Distribution and Economic Transition M.E. Sharpe: A rmonk, N.Y.
Sen, A.K. (1983), "Poor, Relatively Speaking", Oxford Economic Papers , 35(2): 153-
169.
Solow, Robert (1956) "A Contribution to the Theory of Economic Growth" Quarterly
Journal of Economics . 70(1): 65-94.

Song, Lina (2000) “Diversification of Household Production in Rural China:
Determinants and Outcomes" in L. West and Y. Zhao (ed.) Rural Labor Flows in
China , University of California Press, Berkeley.
Wang QB, Shi GM and Zheng Y (2002) “Changes in income inequality and welfare under economic transition: evidence from urban China” Applied Economics Letters 9
(15): 989-991.
Wu. F. (2004), “Urban poverty and marginalization under market transition: the case of Chinese cities”
International Journal of Urban and Regional Research , 28(2): 401-
423.
Wong, Linda (1998), Marginalization and Social Welfare in China, Routledge:
London.
Wood, Adrian (1995), North-South Trade, Employment, and Inequality: Changing
Fortunes in a Skill-Driven World , Oxford University Press: Oxford.
World Bank (2001), World Development Report 2000/01: Attacking Poverty Oxford
University Press: New York.

Table 1: Basic information on household income per capita, Household Income
Project Surveys
Income per capita
Median income per capita
% of income from:
1. Earnings of working members
2. Income of retired
1988
4820
4268
43.2% 59.4% 51.7% 60.2%
8.0%
0.5%
1995
6673
5365
13.5%
0
1999
8771
7180
18.3%
0
2002
9853
8365
16.7%
0 3. Income of other non-working members
4. Income from private or individual enterprises
5. Income from property
6. Miscellaneous income (including private transfers and special income)
7. Subsidies less taxes (except housing subsidy and coupons) and income in kind
0.7%
0.5%
4.45%
15.3%
0.5%
1.3%
3.4%
1.2%
1.8%
0.9%
2.2%
0.2%
2.7%
0.6%
3.1%
0.2%
8. Ration coupon subsidy
9. Housing subsidy
10. Rental value of owner occupied housing
5.2%
18.4%
3.7%
Number of observations 9005
Note: yuan per year, 2002 constant prices.
0
10.1%
10.8% 18.5% 13.9%
6929
0
6.5%
3998
0
2.8%
6835

Table 2: Growth rates of household income, Household Income Project and SSB data compared
Household income project State Statistical Bureau
(SSB)
1988-1995
1995-1999
1999-2002
1988-2002
4.65%
6.83%
3.88%
5.11%
6.63%
5.42%
8.98%
6.79%
Table 3: Real income per capita by decile, Household Income Project Surveys
10 th
20 th
30 th
40 th
50 th
60 th
70 th
80 th
90 th
1988
2705
3180
3553
3902
4268
4663
5186
6019
7477
1995
2855
3542
4142
4709
5365
6080
7040
8545
11489
1999
3502
4504
5400
6284
7180
8159
9540
11506
15030
2002
4024
5134
6203
7271
8365
9535
11035
13380
17211
Note: yuan per year, 2002 constant prices
Table 4: Inequality in household income, 1988-1995
Gini coefficient
Atkinson index (1.5)
Generalised entropy (1) or Theil’s
T index
Generalised entropy (2) or Theil’s
L index
1988
0.235
0.123
0.100
0.091
1995
0.328
0.238
0.223
0.184
1999
0.331
0.244
0.262
0.193
2002
0.318
0.225
0.172
0.170

Table 5: Poverty indices, with $2 and $3 a day poverty lines
With $2 a day poverty line
Head count, P0
Poverty gap, P1
Squared poverty gap, P2
With $3 a day poverty line
Head count, P0
Poverty gap, P1
Squared poverty gap, P2
With half median income poverty line
Head count, P0
Poverty gap, P1
Squared poverty gap, P2
1988
7.33%
1.17%
0.31%
36.36%
7.50%
2.34%
3.8%
0.59%
0.15%
1995
7.00%
1.64%
0.62%
23.81%
5.88%
2.28%
9.3%
2.25%
0.86%
1999
3.66%
0.92%
0.38%
12.39% 8.52%
3.12% 1.26%
1.26%
11.8%
3.00%
1.21%
2002
2.08%
0.38%
0.17%
0.72%
12.8%
3.08%
0.12%
Note: Poverty lines are in constant 1985 PPP dollars. One 1985 PPP dollar corresponds to 1212 yuan in 2002 prices.
Table 6: Decomposition of poverty changes into growth and redistributional components (headcount poverty index)
Residual
(a) $2 a day poverty line
Growth component
Redistribution component
Total change in poverty
1988-1995
1995-1999
1999-2002
1988-2002
(b) $3 a day poverty line
1988-1995
1995-1999
1999-2002
1988-2002
-6.02%
-4.09%
-0.96%
-7.19%
-25.87%
-13.42%
-3.35%
-34.95%
11.64%
1.43%
-0.35%
13.05%
12.01%
1.25%
-0.47%
7.96%
-5.95%
-0.67%
-0.28%
-11.11%
1.32%
0.74%
-0.04%
-0.88%
-0.33%
-3.34%
-1.58%
-5.25%
-12.54%
-11.42%
-3.87%
-27.83%

Table 7: Decomposition of poverty changes by employment status of household head, 1995-2000 a) $2 a day poverty line
Employed
Unemployed
Retired
Other non-
1995
Population Poverty
Share headcount
80.31
0.4
18.37
7.02
31.03
5.85
2002
Population Poverty
Share headcount
71.04
6.19
22.04
1.86
6.97
1.17
Population shifts
-0.65075
1.796637
0.214695
Contribution of
Intra-group changes Interaction
-4.144 0.478332
-0.09624 -1.39307
-0.85972 -0.17176 participants
Sum
0.93
100
17.91
7.00
0.73
100
9.33 -0.03582
2.08 1.324758
-0.07979 0.01716
-5.17975 -1.06934 b) $3 a day poverty line
Employed
Unemployed
Retired
Other non-
1995
Population
Share
Poverty headcount
80.31
0.4
18.37
24.07
50.58
21.46
2002
Population
Share
Poverty headcount
71.04
6.19
22.04
7.83
21.85
6.36
Population shifts
-2.23129
2.928582
0.787582 participants
Sum
0.93
100
36.32
23.81
0.73
100
28.67 -0.07264
8.52 1.412235
Notes: all numbers are percentages. Poverty lines defined in $PPP 1985 prices.
Contribution of
Intra-group changes Interaction
-13.0423 1.505448
-0.11492
-2.77387
-0.07115
-16.0023
-1.66347
-0.55417
0.0153
-0.69689

8%
7%
6%
5%
4%
3%
2%
1%
0%
10th 20th 30th 40th 50th 60th 70th 80th 90th
Percentile point
88-95 95-99 99-02 88-02

Figure 2: Poverty incidence curves 1988-2002
40
37.5
35
32.5
30
27.5
25
22.5
20
17.5
15
12.5
10
7.5
5
2.5
0
60
57.5
55
52.5
50
47.5
45
42.5
10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50
Poverty line (100 yuan per year, 2002 prices)
1988 1995 1999 2002