Home, school, and community deprivations: A multi-context approach to childhood poverty in China

Abstract

Few studies of childhood poverty in China have considered social deprivations, and fewer still have considered deprivations in contexts beyond the household. In this article, the authors propose a multi-context poverty measure that includes economic and social deprivations in the family, school, and community domains. The authors compare this measure to standard money-metric measures of poverty using a 15-year longitudinal study of children from 100 villages in Northwest China. Nearly a quarter of multi-context poor children are not income poor, while almost three-fourths of the income poor are not multi-context poor. Social and community deprivations are only weakly associated with “money-metric” deprivations but are significantly linked with short and long-term welfare outcomes for children. The multi-context approach can be tailored to index household and community poverty alleviation targets.

Introduction

The World Bank estimates that China has lifted 800 million people out of poverty since 1978.¹ Despite this remarkable success story, the scale of poverty that remains is immense: In a white paper issued in 2016,² the State Council reported that 55.75 million people still live in poverty, with the majority of these in extreme poverty. A World Bank estimate for 2015 suggested that about 7 percent of the world’s poor were in China.³ Addressing this problem has reached the top of the national policy agenda. President Xi Jinping has made alleviating rural poverty a signature issue, pledging in 2015 that China would lift those living under the poverty line out of poverty by 2020.⁴ This pledge was reaffirmed in a report delivered by Xi Jinping to the 19th National Congress of the Communist Party in 2017.⁵

Children are recognized as one of the specific groups vulnerable to poverty.⁶ The World Bank estimated that in 2003, 16.3 percent of boys and 17.3 percent of girls under age 16 were consumption poor, compared to 13.1 percent for the entire population.⁷ Yet, estimates of the scale of childhood poverty are relatively scant, have been characterized by little agreement on a standard definition, and, until recently, have primarily adopted a conceptualization tied to household income or economic resources.⁸ Significantly, while Xi’s statement refers explicitly to an economic poverty line to identify the rural poor, the commitment to poverty reduction implies a conceptualization of poverty that is both multi-dimensional and multi-contextual. For example, a white paper issued by the State Council⁹ refers specifically to multiple deprivations targeted by the new poverty alleviation initiatives and recognizes the importance of the community context: “By 2020, the State is committed to ensuring that the impoverished rural population has stable access to adequate food and clothing, compulsory education, and basic medical services and housing; to realizing a growth rate of per capita disposable income in poor rural areas higher than the national average; to achieving indices of major basic public services close to national average levels; to ensuring that the rural population living below the current poverty threshold and all impoverished counties are all lifted out of poverty; and to solving the problems of regional poverty.” Similarly, the World Bank’s last poverty assessment report for China recognized that the reasons for poverty were diverse and suggested the need for programs tailored to specific household and community needs. ¹⁰

In recent decades, most of the work on household poverty in China has focused on economic poverty—income or consumption poverty—by adopting a World Bank threshold or the national poverty line to distinguish poor and non-poor households (see, for example, World Bank, 2009). ¹¹ An alternative, multidimensional poverty measurement approach¹² has emerged that focuses on deprivations in various household domains rather than just income or consumption. A number of studies have applied such poverty approaches in China.¹³ For example, using data from the China Family Panel Studies (CFPS) fielded in 2012, Alkire and Shen applied Global Multidimensional Poverty Index (MPI) estimates and documented high levels of mismatch at the household level between monetary poverty and multidimensional poverty. The paper found higher levels of multidimensional poverty in Western China than in Eastern or Central China, and in rural than in urban areas.¹⁴ Some studies have focused on children and have found falling poverty but persisting regional gaps; some regions are characterized by severe deprivations.¹⁵

This article argues that money-metric household poverty measures—based on household income or consumption alone—miss important deprivations in other domains of children’s lives. These domains include social domains within the household—a point recognized, to some degree, in existing multidimensional approaches. However, domains beyond the household, in particular, the child’s school and the community, are also important contexts of child development, and these domains are not considered in the existing literature in China.¹⁶

A multi-context poverty measure that includes indicators of economic and social deprivations across the key contexts of child development is proposed. This measure is compared to standard money-metric measures of poverty using a 15-year longitudinal study of children from 100 villages in Northwest China. This exercise shows that social and contextual deprivations are weakly associated with “money-metric” household deprivations, but are significantly associated with short and long-term welfare outcomes for children.¹⁷ Finally, the article suggests that the proposed approach is appealing in that it can be tailored to index specific community- and household poverty alleviation targets, as China embarks on an ambitious new poverty alleviation initiative.

Framework

Conceptualizing childhood household poverty

The article begins by presenting a framework for measuring childhood poverty. Childhood poverty encompasses direct economic deprivation, but also other disadvantages at home and in a child’s broader ecological environment in the school and community.¹⁸ Household poverty is an established risk factor for many outcomes that affect capacity to function in school, including physical health and language and cognitive development, though specific causal processes are not yet well understood.¹⁹ Child development scholars have variously conceptualized the impact of family socioeconomic status on child development as having to do with environmental resources and psychological influences; family investment and family stress; and financial resources and family process/parental socialization.²⁰ While these models differ in their specific categorizations of influence, they have in common a conceptualization of family socioeconomic status as shaping children’s outcomes through at least two key mechanisms: first, shaping resources and investments, and second, protecting from material deprivations and stressors.²¹

At a basic level, a family’s standard of living, including adequate food, housing, clothing, and medical care can shape children’s readiness to learn.²² Poor children in rural China face elevated risks of being chronically undernourished, living in food insecure households, and lacking access to vision correction.²³ Poor children tend to be exposed to more “health insults” than their more affluent counterparts, and in China, the lowest socioeconomic status children are more likely than others to report having experienced serious illnesses.²⁴ The “toxic stress” associated with child poverty—experiences of severe, uncontrollable, and chronic adversity—can strain children’s capacity for behavioral functioning. ²⁵ In poor parts of rural China in the early 2000s, for example, poor children were likely to be in families with parents who were working to the point of exhaustion and injury—and possibly migrating away from home for this purpose—to cover school fees. These children may have parents who are ill with no access to health care, parents who face crushing debt, parents who are stressed to the breaking point by struggles of daily life, and parents with no capacity to advise and help with educational problems, due to their own limited education. Nationally, lower socioeconomic status children face heightened likelihood of living apart from parents and their parents are less available for regular homework supervision.²⁶ Poor rural children of migrant parents commonly reside with grandparents who are often illiterate and have limited experience with the public education system. Cumulative adversity in exposure to stressful life events has been linked to more behavioral problems among children.²⁷

Broader ecological contexts

In the child development literature, it is well-recognized that beyond the family, the broader community and school contexts experienced by poor children may be unlike those experienced by the non-poor. Further, these contexts tend to be linked because of the geographic basis of school catchments. Many mechanisms of community effects have been put forward.²⁸ These range from basic hard and soft infrastructure to spatial mismatch with job opportunities to socialization and peer networks.²⁹ However, few studies have focused on the particular issues that might be important to youth opportunity in poor rural communities in developing country contexts. Earlier research in rural China has linked village income levels to enrollment and persistence, net of family characteristics, and has demonstrated substantial math achievement differences across communities, associated in part with the economic and social resources in those communities.³⁰

The quality of school resources is also linked to community wealth in many countries. For example, in the United States, children living in poverty tend to be concentrated in schools with inadequate resources and poorly skilled teachers.³¹ Per-pupil expenditures vary tremendously, even within districts. ³² Many studies have debated whether and how school resources matter in the United States³³ and in developing and industrialized countries.³⁴ But cross-national work has suggested that whether or not children stay in school at all may be linked to quality: Huisman and Smits’ multinational studies have found that enrollment decisions were associated with quality-related characteristics of the available educational facilities in districts, such as number of teachers.³⁵ Research using a multi-province household survey implemented in western China using data from 2011 suggested that boy’s enrollment, more than girls’, was sensitive to school facility quality.³⁶

Implications

Childhood poverty encompasses a set of household conditions—economic and social—that may impede a child’s capacity to function optimally in the classroom. At the same time, children in poor families often also reside in poor communities, where resource deprivation can constrain the supply and quality of schooling. Yet, studies of childhood poverty in China, while beginning to apply multi-dimensional approaches, have not considered the domains outside the family in which children function. This article proposes a multi-context approach to measuring childhood poverty that encompasses social as well as economic deprivations, in the community and school as well as in the household. The remainder of this article compares this new approach—in terms of who is identified as poor and in terms of associations with child welfare outcomes—to standard income- and consumption-based approaches. This article will draw on a case study of Gansu Province—one of China’s poorest provinces—to analyze childhood poverty.

Data and background

Field site

Gansu Province is located in China’s arid Northwest. The province is long and narrow, stretching across desert, mountainous and hilly areas, and vast grasslands. In 2010, the most recent census year, Gansu’s population was 25.6 million people.³⁷ This figure would make Gansu the 47^th largest country in the world (in population size) in that year—the province is about the size of Ghana or North Korea.³⁸

Gansu is a useful site for studying childhood poverty in rural China. First, Gansu remains a majority rural province: in 2010, the most recent census year, 64 percent of the population resided in rural areas.³⁹ Second, Gansu is one of China’s poorest provinces: in 2010, it ranked last among provinces in per capita rural household income and third to last in per capita GDP.⁴⁰ Per capita GDP in 2010 was $4,031 USD—about equivalent to the 2010 figure for Iraq.⁴¹

Data source

This article analyzes data from the Gansu Survey of Children and Families (GSCF), a longitudinal study of 2,000 children in rural villages conducted in four waves: 2000, 2004, 2009 and 2015. The sampled children lived in 100 villages located in 20 different counties in Gansu Province. They were aged 9 to 12 when first interviewed in the year 2000 (GSCF-1). The same children were re-interviewed in 2004 when they were 13 to 16 (GSCF-2), again in 2009 when they were 18 to 21 (GSCF-3), and finally in 2015, when the children were 24 to 27 years old (GSCF-4). By that time, most had completed their education, and many had made transitions to marriage, migration, and employment.

Key strengths of the GSCF data are as follows. First, detailed questionnaires about household, school and community characteristics allow for comparisons of poverty estimates based on standard income- and consumption-based approaches to estimates based on an extended multi-context poverty approach. Second, the multi-disciplinary approach to measuring child well-being and transitions to adulthood allows for a study of the associations of early poverty exposure with a variety of short-term outcomes related to education and welfare, and long-term educational attainment in young adulthood.

Defining the poor

Traditional approach: “money metric” measures of poverty based on income or consumption

In most of the previous work in China, income- and/or consumption-based measures are used to identify whether or not a household is poor. Both are defined as occurring when a family’s income or consumption per capita falls below an established threshold. Typically, it is measured with respect to families and not the individual. The international standard for extreme poverty was originally set by the World Bank to be $1 a day per person in 1996 US prices. Newer global poverty estimates recently released by the World Bank use an updated international poverty line of $1.90 a day in 2011 prices using purchasing-power parity exchange rates. According to this standard, China’s poverty rate fell from 88 percent in 1981 to 6.5 percent in 2012.⁴² The percentage of the population living below the international poverty line continued to fall to 4.1 percent in 2014.⁴³

A multi-context, multi-dimensional poverty approach

Deprivations can occur in different domains of life, and it is unclear whether income or consumption-based measures fully account for these deprivations. This article defines a multi-context (MC) poverty headcount ratio inspired by the approach developed by Alkire and Santos (hereafter A-S)⁴⁴ for the Human Development Report in 2010, which complements “money metric” measures of poverty by considering overlapping deprivations suffered in other dimensions at the same time. A-S⁴⁵ use micro data from household surveys to measure multiple deprivations at the household level in education, health, and standard of living.⁴⁶ Their measure aims to capture deprivations of individuals of all ages and includes individual education and health outcomes. The measure proposed here aims to capture the environment in which children grow up, which is typically beyond their control and strongly influences their development, so does not include child outcomes directly. Further, the proposed measure extends the multi-dimensional approach to include two key contexts of children’s lives beyond the household: the school and the community.

Following the A-S methodology,⁴⁷ this article proposes a multi-context poverty measure for Gansu children that includes domains in the household beyond economic well-being. The measure also includes domains that go beyond the family. Specifically, for each major environment in which children grow up—the family, school, and community, two kinds of deprivations are considered: economic and social. Thus, in total, six domains are considered: family economic, family social, school economic, school social, village economic, and village social. Table 1 lists the detailed indicators included for each of these domains.

Table 1.

Poverty Indicators in Six Domains

Description	Components	Cutoffs	Share of the population derpived in each dimension
1. Family economic domain
Financial resources and living standards.	consumption per capita	using China low income level in 2000, deprived if annual consumption per capita<=865 RMB	0.253
	wealth per capita	deprived if annual wealth per capita is lower than the 25th percentile (which is 1342 RMB) of the 2000 sampled families	0.250
	income per capita	using China low income level in 2000, deprived if annual income per capita<=865 RMB	0.508
	time to get water (unit: minutes)	deprived if the time to get clean water is higher than the 75th percentile of the 2000 sampled families, which is 20 minutes per trip	0.199
	home construction materials (bricks or concrete =1, wood or others=0)	deprived if home construction materials is not bricks or concrete	0.151
	cooking fuel (coal or gas=1, wood or other=0)	not deprived if cooking fuel is coal or natural gas; deprived if cooking fuel is firewood, straw or others	0.371
2. Family social domain
Human and social capital.	mother's literacy	deprived if mother is reading and writing illiterate	0.495
	father's literacy	deprived if father is reading and writing illiterate	0.199
	mother's health status	deprived if mother has poor health, or disabled, or passed away	0.105
	father's health status	deprived if father has poor health, or disabled, or passed away	0.068
	parental migration	deprived if either father or mother is a migrant worker	0.284
	maternal life satisfaction scale	construct a maternal life satisfaction scale from a set of mother's self-response to questions like "are you proud of yourself?" (see Appendix for more details on how we define mother's life satisfaction scale); define deprived if below the 25th percentile (i.e., -0.123)	0.250
3. School economic domain
Financial resources and basic infrastructure standards.	per capita student expenditures	deprived if student expenditure per capita is below the 25th percentile (i.e., less than 5.46 RMB) of that of all schools in the sample	0.241
	with unsuitable classrooms	deprived if the school has unsuitable classrooms, unsuitable rooms = dangerous classrooms, non-rainproof classrooms, or students without enough chairs/desks	0.248
	school with a library?(yes=1, no=0)	deprived if school has no libraries	0.288
Human capital and organizational resources/environment.	%teachers' education degree higher than zhongzhuan	compute the percentage of teachers whose education degrees are lower than zhongzhuan (i.e., middle school ungraduated, high school ungraduated, zhongzhuan ungraduated, or others); define deprived if the school average percentage is above 75th percentile, which is 100%	0.235
	teacher turnover rate	the percentage of teachers leaving and coming is above the 75th percentile, which is 0.25	0.186
	total number of teachers	total number of teachers is below the 25th percentile among all schools, which is 7 teachers per school	0.241
	teacher time for teaching related activities (unit: hours)	compute the school average of teachers' total hours spent on course preparation, attending teaching-related seminars and extracurricular further education; define deprived if the average time is below the 25th percentile, which is 18 hours per week	0.244
	teacher morale	construct a job satisfaction scale from teacher's response to a set of questions like "do you like to be a teacher?" for each teacher and then compute the school average; define deprived if the school average scale is below the 25th percentile, which is -0.092	0.245
	teacher professional development opportunities	construct teacher's professional improvement scale from a set of questions like "do you attend any professional work service organized by schools?", compute the school average for all teachers, define deprived if the school average is below the 25th percentile among all schools, which is -0.127	0.249
	teacher collaboration (preparing lectures together more than half the time? Yes=1, no=0)	deprived if teachers preparing lectures together less than half the time	0.254
	school average absence rate among all grades	the school average of the student absence rate among all graders is above the 75th percentile, which is 2.59%	0.243
5. Village economic domin
Financial resources and basic infrastructure standards.	income per capita (RMB yuan)	village averge income per capita is below the 25th percentile of all villages' averages, which is 619.23 RMB	0.250
	land per capita (unit: mu)	village-level land per capita is below the 25th percentile of all villages' averages, which is 1.486 mu per capita	0.242
	electricity access (% households with electricity)	village average electricity access rate is below the 25th percentile among all villages, which is electricity access rate lower than 95%	0.240
	village has preschool/kindergarten classes (yes=1, no=0)	deprived if no preschool/kindergarten in the village	0.242
	village has primary school (yes=1, no=0)	deprived if no primary school in the village	0.290
	village or nearby has middle school (yes=1, no=0)	deprived if no middle school in the village	0.200
	village is mainly mountainous areas? (yes=1, no=0)	deprived if the village is mainly mountainous areas	0.300
	distance to nearest township (kilometers)	deprived if the distance from the village to the nearest township is above the 75th percentile, which is 7 kilometers	0.200
	distance to county seat (kilometers)	deprived if the distance from the village to the nearest county seat is above the 75th percentile, which is 32 kilometers	0.250
	village with railway/highway passing through (yes=1,no=0)	deprived if without railway/highway passing through	0.460
	village with bus stations (yes=1,no=0)	deprived if without bus stations	0.410
	village without telephone, mail and radio service (yes=1,no=0)	deprived is without telephone, mail or radio service	0.090
6. Village social domain
Human capital and organizational resources/environment.	illiteracy rates among female labor force (%)	deprived if the village average illiteracy rate among female labor force is above the 75th percentile, which is 30.77%	0.242
	illiteracy rates among male labor force (%)	deprived if the village average illiteracy rate among male labor force is above the 75th percentile, which is 47.06%	0.232
	how often village children fight violently in groups (always/sometimes=1, never=0)	deprived if group fighting among village children is prevalent	0.505

Source: Authors’ calculations, Gansu Survey of Children and Families, 2000.

Steps to Define Multi-Context Poverty:

⁽¹⁾For each poverty indicator, we define the poverty cutoff as listed in Column C of the above table.

⁽²⁾A household is deprived in a domain (family economic, family social, school economic, school social, village economic or village social), if it is deprived in half of the poverty indicators within that domain.

Domain 1, the family economic domain, consists of 6 indicators of household financial resources and living standards: consumption per capita, wealth per capita, income per capita, time to get water, home construction materials, and cooking fuel. Domain 2, the family social domain, consists of 6 indicators reflecting family human and social capital, including mother’s literacy, father’s literacy, mother’s health status, father’s health status, parents’ migration, and maternal life satisfaction. Domain 3, the school economic domain, considers 3 indicators of village school economic resources: per capita student expenditure, percentage of unsuitable classrooms (dangerous or non-rainproof classroom, or students without enough chairs/desks), and whether the school has a library. Domain 4, the school social domain, includes 8 variables that reflect teachers’ quality and collaboration within the school. Domain 5, the village economic domain, includes 12 indicators of the villages’ infrastructure and access to the outside world. Domain 6, the village social domain is comprised of the village average education level and the prevalence of fighting among groups of children in the village.

Within each of the above six domains, a multi-context (MC) headcount poverty indicator is defined using a dual cut-off method: first define cut-offs () for each poverty indicator (), assign weights () to these indicators, and then calculate an aggregate deprivation measure and apply a poverty cut-off (k). The MC poverty headcount indicator is constructed in two stages. First, within each domain, each indicator is weighed equally to construct a domain-specific MC poverty headcount measure. A cut-off of 50 percent, which is equivalent to one half of the weighted indicators, is used as the poverty threshold for each domain. Next, the six domain-specific poverty headcount indicators are used to construct the overall MC poverty indicator, weighing each of the six dimensions equally. Again, a cross- dimensional cut-off of 50 percent is applied, so a child is categorized as multi-context poor if he or she is deprived in at least 3 of the 6 domains.

For example, to identify whether a person is deprived in the family economic domain, this approach first defines cut-offs for all six indicators in that domain. The deprivation cut-offs in the year 2000 are listed in column (3) of Table 1. A household is deprived in consumption if the annual consumption per capita is below the 2000 national poverty line of 865 RMB. Similarly, a household is defined as deprived in income if its annual income per capita is lower than 865 RMB. Also included are living standards in the family economic domain: a household is deprived in housing if the home construction materials are not bricks or concrete; a household is deprived in utility if the cooking fuel is not coal or gas. Since some indicators do not have clear thresholds between deprived and non-deprived, the 25th percentile is used as the cut-off in those cases, which yields the same rate of deprivation as for consumption poverty--the poverty measure preferred by economists. For example, a household is coded as deprived in wealth if its wealth level per capita is below the 25th percentile among the 2000 families of the sampled children. Similarly, a household is deprived in water access if the time to get clean water is above the 75th percentile.

A deprivation score is assigned to each of these six indicators. The deprivation score is the product of two measures: the condition of being deprived or not and the weight of the indicator. These weights are set to be equal, so that each of the 6 indicators in the family economic category is worth 16.7 percent. Then, a child is multi-context poor in the family economic domain if its weighted deprivations sum up to 0.5 or more.

As this approach assigns equal weights to the indicators within a domain, when the weighted deprivation score is greater than 1/2, this is equivalent to saying that the household is deprived in at least half of the poverty indicators. This can be written as:

$$\begin{array}{l}{\text{Q}}_i=1\text{if}\sum _j{w}_j^{*}1\left(y_{ij}<z_j\right)\geqq 1/2,\\ =0\text{otherwise,}\end{array}$$

where y_ij is the level of individual i in indicator j, z_j is the cut-off for poverty indicator j and w_j is the weight of poverty indicator j. Q_i measures whether individual i is poor or not.

The headcount ratio, H, is the proportion of poor in the population:

$$H=\frac{q}{n}=\frac{\sum _i{Q}_i}{n}$$

where q is the number of people who are poor in the family economic domain and n is the total population.

Similarly, indicators of deprivation are defined in the remaining five domains. Within each of these domains, the indicator is set equal to one if the household is deprived in at least half of the indicators. Then, a similar exercise is conducted using the constructed deprivation measures for each of the six domains. If a child is deprived in at least three of the six domains, he or she is categorized as being multi-context poor.

An additional measure is also presented, defined as the multi-context poverty index (MCI) and based on the multidimensional poverty index (MPI) or “adjusted headcount ratio” developed by A-S. The MCI combines information on the prevalence of poverty and the average depth of a poor person’s deprivation. The depth of deprivation, A, reflects the proportion of the weighted component indicators in which, on average, poor people are deprived. For multi-context poor households, the deprivation scores are summed and divided by the total number of poor people to obtain the average depth of deprivation A: $A=\frac{\sum _i{c}_i}{q}$ for all i who are poor.

Here, c_i is the deprivation score that the ith poor individual experiences.$c_i=\sum _j{w}_j\ast 1(y_{ij}<z_j)$

The MCI value—the adjusted headcount ratio— is “adjusted” by the depth of deprivation, or the intensity of poverty. Specifically, it is the product of the multidimensional poverty headcount ratio H and the average breadth of deprivation A: MCI=H*A. As Santos and Alkire explain,⁴⁸ if a society were 80 percent poor people, and all of them were deprived in all the indicators, then A would equal 1 and the MCI would equal H. In contrast, if there were a society where 100 percent of people were poor, then the MCI would be equal to A. This adjusted headcount ratio or MCI can be interpreted as the proportion of weighted deprivations that the poor experience in a society out of the total potential deprivations that the society could experience.⁴⁹ It can be calculated for each of the six domains, or for overall multi-context poverty using the six domain-specific poverty indicators.

Results

Child multi-context poverty in Gansu

Table 2 shows overall and domain-specific multi-context poverty headcounts H, as well as the MCI or adjusted headcount ratio. Overall, 18 percent of households are multi-context poor (H=0.180). The 18 percent of households that are poor are deprived in 59 percent of indicators. And the MCI score indicates that the poor experience about 10.6 percent of total potential deprivations that could be experienced.

Table 2.

Multi-context poverty headcount ratio (H), breadth of deprivation (A) and MCI, by domain and overall

	Headcount ratio (H)	Average breadth of deprivation (A)	Adjusted headcount/MCI (H*A)
(1) Family economic	0.275	0.604	0.166
(2) Family social	0.170	0.553	0.094
(3) School economic	0.190	0.753	0.143
(4) School social	0.150	0.542	0.081
(5) Village economic	0.160	0.581	0.093
(6) Village social	0.230	0.768	0.177
All domains	0.180	0.591	0.106

Note: To compute the poverty headcount ratio and MCI across all dimensions, we define being poor as those who are deprived in at least 3 of the 6 dimensions.

Source: Authors’ calculations, Gansu Survey of Children and Families, 2000.

Focusing first on family deprivations, around 27.5 percent of children in the sample are deprived in at least half of the family economic measures and about 17 percent are deprived in at least half of the family social measures. Striking in these domains are that half of children are living in income poverty; half have mothers who are illiterate; and over a fourth have a parent who is a migrant worker. One in ten have mothers with poor health, disabilities, or who have passed away, and over one in twenty have a father in this situation.

Results show that 19 percent of children are deprived in at least half of the school economic domains and about 15 percent are deprived in the school social domains. For example, 28.9 percent of students attend schools with no library. About 16 percent of children were deprived in at least half of village economic domains. With regard to specific indicators, 24.2 percent of children live in villages with no preschool or kindergarten, 29.0 percent live in villages with no primary schools, 30.0 percent live in mountainous villages, 41.0 percent live in villages with no bus stops, and 9.0 percent live in villages with no telephone, mail, or radio service. About 23 percent of children were deprived in at least half of village social domains. For example, half of children live in villages where children sometimes or often fight in groups.

Associations of household “money-metric” deprivations and other deprivations

Table 3 describes the correlation of deprivations across different domains. Panel A reveals generally weak associations of family economic deprivations with deprivations in other domains, and very weak correlations of family social deprivations with other domains. There is a moderate association of school social deprivation with village social deprivation. Panel B shows the probability of being deprived in the column domains, conditional on deprivations in row domains. Among children deprived in the family economic domain, there is a 29 percent chance of poverty in family social and school economic domains, a 24 percent chance of poverty in school social domains, a 42 percent chance of living in an economically deprived village, and a 38 percent chance of living in a socially deprived village. A child in an economically deprived village has a 56 percent chance of being in an economically deprived household, a 30 percent chance of being in a socially deprived household, a 40 percent chance of being in an economically deprived school, a 34 percent chance of being in a socially deprived school, and a 48 percent chance of being in a socially deprived village. These patterns illustrate the multiple deprivations that a child can experience, but also that deprivations in different domains are by no means reducible to one another.

Table 3.

Associations of deprivations in different domains

Domain	(1) Family	(2) Family	(3) School	(4) School	(5) Village	(6) Village
	economic	social	economic	social	economic	social
Panel A. Correlations of deprivation in each domain
(1) Family economic	1
(2) Family social	0.1327	1
(3) School economic	0.1499	0.0862	1
(4) School social	0.2242	0.0969	0.2249	1
(5) Village economic	0.2749	0.1329	0.2058	0.1986	1
(6) Village social	0.2062	0.0626	0.0382	0.5024	0.28	1
Panel B. Probability of being deprived in the column domain, conditional on deprivation in row domain
(1) Family economic	1.000	0.292	0.290	0.235	0.416	0.380
(2) Family social	0.374	1.000	0.255	0.166	0.281	0.241
(3) School economic	0.397	0.269	1.000	0.253	0.411	0.272
(4) School social	0.446	0.245	0.346	1.000	0.468	0.859
(5) Village economic	0.562	0.298	0.404	0.337	1.000	0.475
(6) Village social	0.380	0.189	0.196	0.444	0.341	1.000

Note: Each cell in panel (b) reports the probability of being deprived in each column domain conditional on being deprived in the row domain.

Source: Authors’ calculations, Gansu Survey of Children and Families, 2000.

Identification of poor children: Money-metric and multi-context approaches

Table 4 shows cross-tabulations of children identified as poor by the most commonly used monetary approaches, compared to the multi-context approach. Consumption per capita, or income per capita if consumption data is not available, is used by the World Bank to make global poverty estimates. When applying the national low-income line (865 RMB in 2000) to the GSCF dataset, 50.8 percent of people are income poor and 25.3 percent of households are consumption poor. One might expect that most of the MC poor children—18 percent of the full sample–would also be income/consumption poor, because the level of income and consumption poverty is higher than MC poverty. However, this is not the case. Nearly a quarter of the MC poor children are not income poor (4.2 percent of the full sample), and about half are not consumption poor (8.95 percent of the full sample). Conversely, almost three-fourths of the income poor are not MC poor (37.02 percent of the full sample) and about two-thirds of the consumption poor are not (16.31 percent of the full sample). Thus, traditional money-metric definitions of poverty do not reduce to multi-context poverty, and vice-versa.

Table 4.

Cross-tabulation of poverty status designations using standard “monetary” poverty measures and multi-context poverty measures

Panel A. MC and Income Poverty
	Not MC Poor	MC Poor	Total
Not Income Poor	45.03	4.20	49.23
Income Poor	37.02	13.75	50.77
Total	82.05	17.95	100.00
Panel B. MC and Consumption Poverty
	Not MC Poor	MC Poor	Total
Not Consumption Poor	65.74	8.95	74.69
Consumption Poor	16.31	9.00	25.31
Total	82.05	17.95	100.00

Note: Distribution of the child sample who are non-poor or poor by type of poverty measure. Numbers are in percentages.

Source: Authors’ calculations, Gansu Survey of Children and Families, 2000.

Table 5 shows that the correlations of childhood multi-context poverty with traditional income and consumption poverty measures are weak, at 0.24 and 0.27, respectively. The correlations of money-metric poverty measures with the family economic domain of poverty are greater (0.492 for income and 0.548 for consumption), which is to be expected given the inclusion of income and consumption as indicators in this domain. Significantly, Table 5 shows very weak correlations between traditional money-metric poverty measures and the social domains of deprivation, and most of the school and community domains of deprivation.

Table 5.

Correlation of multi-context poverty measures and standard “monetary” poverty measures

	Income Poverty	Consumption Poverty
Multi-context poverty	0.2417	0.2671
Deprived in:
Family economic domain	0.492	0.548
Family social domain	0.078	0.043
School economic domain	0.107	0.041
School social domain	0.097	0.123
Village economic domain	0.212	0.154
Village social domain	0.094	0.171

Source: Authors’ calculations, Gansu Survey of Children and Families, 2000.

Poverty gaps in short-term welfare outcomes, by poverty definition

Figures 1 and 2 show gaps in child development outcomes associated with standard monetary poverty measures and the new multi-context measure (Figure 1) as well as with specific domains of deprivation (Figure 2). The full tables upon which figures are based, including all indicators, are contained in Appendix Tables 1 and 2. Figure 1 indicates that, in this sample, the gaps in contemporaneous child development outcomes associated with multi-context poverty tend to be greater than corresponding gaps associated with standard monetary measures. Figures 2a and 2b show significant gaps in school math and language performance associated with each type of deprivation, with the exception of village social deprivations and math achievement. Figures 2c and 2d show that internalizing and externalizing behavioral problem gaps were associated with deprivations in each domain save the household social domain. Findings are consistent with previous research that has indicated that children residing in disadvantaged communities exhibit greater levels of internalizing and externalizing behaviors, above and beyond individual characteristics, and family contexts.⁵⁰ Overall, these patterns indicate that deprivations not captured by monetary approaches are significantly linked to welfare outcomes in childhood.⁵¹

Figure 1:

Short- and Long-term Outcome Gaps Associated with Poverty by Poverty Measure

Source: Authors’ calculations, Gansu Survey of Children and Families.

Figure 2.

Short- and Long-term Welfare Outcomes by Deprivation Status in Multi-context Poverty Domains

Source: Authors’ calculations, Gansu Survey of Children and Families.

Poverty gaps in long-term welfare outcomes, by poverty definition

Figures 1e and 2e show analogous gaps in longer-term outcomes, measured 15 years later in 2015, focusing on years of schooling attained. Figure 1e displays years of schooling by poverty status, defined by monetary poverty measures and the multi-context poverty measure (1e), and by deprivation in each domain of the multi-context approach (2e). Figure 1e shows that multi-context poverty in childhood is significantly associated with years of schooling 15 years later. Income poverty in childhood and consumption poverty are also associated with years of schooling, but the estimated magnitudes are smaller.⁵² Figure 2e breaks out associations by domain of multi-context poverty. Findings suggest that, in this sample, school economic deprivation has the least association with years of schooling and family economic deprivation has the strongest association with years of schooling. A companion paper⁵³ finds that associations of MC poverty with long-term educational outcomes vary by child gender and sibship structure.

Conclusions

This article has drawn on a 15-year longitudinal study of children in Gansu Province to argue that that poverty measures focused on household income or consumption alone miss important sources of children’s deprivations in other domains. This article has shown, first, that household economic poverty is a poor proxy for deprivations in non-economic domains of home environments. Second, analyses have revealed quite modest associations of household economic poverty with deprivation measures in schools and communities. These domains have not been considered in conventional approaches to studying poverty in childhood in China. Third, and very important, this article has shown that the group of children identified as poor by money-metric approaches and the group identified by a multi-context approach, while overlapping, are not the same. Finally, this article has shown that the proposed multi-context approach is associated with key short-term and long-term welfare outcomes of children.

The domains considered here include social domains within the household—a point recognized in the few existing multi-dimensional studies of childhood poverty—but also schools and communities. A multi-context approach offers a useful complement to income and consumption poverty measures in the investigation of childhood poverty. The survey utilized in this study was designed to capture economic and social resources in children’s lives across contexts and contained a rich set of measures. It is important to acknowledge that some of the specific measures used here to index deprivations are not likely to be replicable using the measures present in more multi-purpose household economic, social, and demographic surveys. However, the general approach of indexing critical social as well as economic deprivations in the household is replicable, as evident in the recent uptick in studies focused on multi-dimensional household poverty in China. Further, the multi-context approach is feasible: many household surveys in China do include community components and could be utilized to incorporate some measures of deprivations at this middle level of context.

Further, the multi-context approach is particularly apt at present, given an ambitious national policy agenda to eliminate rural poverty. The agenda targets multiple deprivations in the household and community:⁵⁴ in 2018, Premier Li Keqiang noted that poverty alleviation efforts would be carried out “through the development of local industries, education, and healthcare” and that the government would tailor measures to individuals and households to ensure that the poor, including the elderly and the disabled, would get the assistance they need.⁵⁵ The multi-context approach can be tailored to this agenda because it is flexible. Deprivation indicators can be aligned directly to poverty alleviation goals at the community and household level and adapted to incorporate the critical deprivations and stressors facing poor rural families, communities, and schools in China.

Appendix.

Definition of Selected Deprivation Indicators

Part A: Scale of Mother’s Life Satisfaction

This article uses mothers’ self-response to the following 13 questions to define their life satisfaction.

G1: Many things you did are meaningful.

G2: You are proud of yourself in many respects.

G3: You already do things well.

G4: Recently, you are often not happy.

G5: What you do often gets praise from others.

G6: You are able to deal with difficulties or problems in life.

G7: Recently, you often feel very lonely.

G8: Neighbours or friends would like to turn to you for help when needed.

G9: Recently, you don’t have a good appetite.

G10: You are confident with your future life.

G11: Recently, you have no interest in anything.

G12: Overall speaking, you feel happy.

G13: Overall speaking, you are very satisfied with your life.

Mothers chose their answers in 4 Likert scales from “strongly disagree”, “disagree”, “agree”, to “strongly agree” for each of the above questions. The values of the 4 scales are reverse coded as needed to make higher values represent higher satisfaction in all questions. Then, STATA command “alpha varlist, std item” is used to generate the index of life satisfaction.

Part B: Scale of Teachers’ Morale

The teachers’ survey, uses the following 5 questions to measure teachers’ morale:

F5. Do you like to be a teacher?

F6. Are you satisfied with the job of the teacher?

F7. Did you obtaining the job willingly or passively?

F8. Is it your ideal job to be a teacher?

F9. Would you like to change careers?

The surveyed teachers answered “yes” or “no” to each of the 5 questions. The authors then used the STATA command “alpha varlist, std item” to generate one index to measure a teacher’s morale in his or her career. The authors compute the school average scale of teachers’ morale.

Appendix Table 1.

Short- and Long-term Outcomes, Children from Poor versus Non-poor Families

	Income poverty			consumption poverty			multi-context poverty
	Poor	Non- poor	Diff.	Poor	Non- poor	Diff.	Poor	Non- poor	Diff.
Standardized Chinese test scores in 2000	−0.125	0.129	0.254***	−0.079	0.026	0.105	−0.574	0.123	0.697***
Standardized Math test scores in 2000	−0.082	0.084	0.167***	−0.098	0.034	0.131*	−0.314	0.071	0.384***
Internalizing problems 2000	0.072	−0.073	−0.145***	0.104	−0.035	−0.139***	0.235	−0.051	−0.286***
Externalizing problems 2000	0.084	−0.086	−0.170***	0.118	−0.040	−0.158***	0.253	−0.055	−0.308***
Years of Schooling	10.699	11.738	1.039***	10.364	11.490	1.126***	9.842	11.482	1.640***

Notes:

^***p<0.01,

^**p<0.05,

^*p<0.1

Source: Authors’ calculations, Gansu Survey of Children and Families, 2000 and 2015.

Appendix Table 2.

Short- and long-term outcome gaps using domain-specific deprivation measures

	Household Economic			Household Social
	Deprived	Non- deprived	Diff.	Deprived	Non- deprived	Diff.
Standardized Chinese test scores in 2000	−0.312	0.113	0.425***	−0.277	0.056	0.333***
Standardized Math test scores in 2000	−0.182	0.072	0.254***	−0.149	0.031	0.180**
Internalizing problems 2000	0.142	−0.054	−0.196***	0.069	−0.014	−0.083
Externalizing problems 2000	0.139	−0.053	−0 191***	0.045	−0.009	−0.054
Years of Schooling	10.098	11.607	1.509***	10.103	11.424	1.320***
	School Economic			School Social
	Deprived	Non-deprived	Diff.	Deprived	Non-deprived	Diff.
Standardized Chinese test scores in 2000	−0.465	0.111	0.576***	−0.403	0.070	0 473***
Standardized Math test scores in 2000	−0.233	0.054	0.287***	−0.168	0.030	0.198**
Internalizing problems 2000	0.181	−0.042	−0.224***	0.206	−0.036	−0.242***
Externalizing problems 2000	0.207	−0.048	−0.255***	0.199	−0.035	−0.234***
Years of Schooling	10.893	11.275	0.382*	10.153	11.377	1.224***
	Village Economic			Village Social
	Deprived	Non-deprived	Diff.	Deprived	Non-deprived	Diff.
Standardized Chinese test scores in 2000	−0.586	0.107	0.693***	−0.284	0.087	0.372***
Standardized Math test scores in 2000	−0.305	0.061	0.365***	−0.071	0.020	0.091
Internalizing problems 2000	0.197	−0.038	−0.235***	0.104	−0.031	−0.135***
Externalizing problems 2000	0.227	−0.043	−0.271***	0.128	−0.038	−0.166***
Years of Schooling	10.114	11.397	1.283***	10.239	11.478	1.239***

Notes:

^***p<0.01,

^**p<0.05,

^*p<0.1

Source: Authors’ calculations, Gansu Survey of Children and Families, 2000 and 2015.