Not a Zero-Sum Game: China’s Internal Migration and the
Wellbeing of Rural-origin Children
Duoduo Xu
PhD Candidate, Center for Applied Social and Economic Research (CASER)
And Division of Social Science
Hong Kong University of Science and Technology
Clear Water Bay, Kowloon
Hong Kong SAR, CHINA
(Email: [email protected])
Jaap Dronkers
Chair International Comparative Research on Educational Performance and Social Inequality
Research Centre for Education and the Labour Market
Maastricht University, the NETHERLANDS
(Email: [email protected])
Xiaogang Wu
Professor, Division of Social Science
Director, Center for Applied Social and Economic Research (CASER)
Hong Kong University of Science and Technology
Clear Water Bay, Kowloon
Hong Kong SAR, CHINA
(Email: [email protected])
Population Studies Center Research Report 16-855
March 2016
We acknowledge support from the Research Grants Council of Hong Kong via the General Research
Fund for the project “Consequences of Internal and Cross-border Migration in China for Children: A
Mainland-Hong Kong Comparison” (646411), granted to the third author (Wu), and the research
facility from the Center for Applied Social and Economic Research, HKUST. Direct all correspondence
to Duoduo Xu (email:[email protected]) or Xiaogang Wu ([email protected]), Center for Applied Social and
Economic Research (CASER), Hong Kong University of Science & Technology, Clear Water Bay,
Kowloon, Hong Kong SAR.
China’s Internal Migration and the Wellbeing of Rural-origin Children
2
Abstract
Two-fifths of the 279 million children in China are directly affected by the on-going massive
internal migration. Using data from a nationally representative survey covering junior high
school students in both rural and urban areas, we examine the causal impacts of different
migration processes (parental migration, child migration, and hukou conversion) on rural
children’s wellbeing measured by cognitive abilities, school engagement, school attachment,
health, educational aspirations, and confidence about the future. Results from propensity
score matching analysis show that migration brings both benefits and costs to children.
Bringing rural children to cities significantly improved their school performance and physical
health on the one hand, but also reduced their educational aspirations and increased their
anxiety towards the future on the other hand. Leaving children behind in the countryside,
while avoiding potential social exclusions in cities, results in a negative impact on their
physical and mental health. Gaining local urban hukou status can improve rural-origin
children’s academic achievement to some extent, but has no effect on the other wellbeing
indicators. These results reveal that the migration process per se and China’s unique hukou
system have generated both opportunities and challenges for children involved.
China’s Internal Migration and the Wellbeing of Rural-origin Children
3
Introduction
Since the 1990s, China has been experiencing the largest population migration in human history.
Hundreds of millions of people have been moving from rural areas to urban areas for better work
opportunities and life prospects. Their migration decisions have also brought transformative and
fundamental impacts on rural families, especially their children. According to a research report
based on the analysis of the 2010 population census, almost 36 million rural migrant children
(aged 17 years or below) lived in cities; one third of all children (i.e., over 61 million) in the
countryside were left behind by migrant parent(s), of which half were left behind by both parents
(ACWF 2013). With such a huge number of children being involved in the migration process, it
is both theoretically interesting and empirically important to investigate the impacts of migration
on the wellbeing of children from rural China.
Two groups of children are directly affected by China’s internal migration process: those
who moved to cities with their parents, and those who were left behind by migrant parents in the
countryside. Despite a surge of interest in the wellbeing of these rural-origin children in recent
years, most studies have yielded inconclusive findings, and the sociological explanations have
been far from clear (Wu and Zhang 2015).
Existing research on the impacts of migration on children suffer from three major limitations.
First, most studies have not developed appropriate and complete comparison strategies.
Following the assimilation theories in the study of international immigration, scholars would
usually compare rural migrants with urban natives. However, this type of comparison reveals
little about the causal impacts of migration per se. From a methodological perspective, as urban
natives are not at risk of migration by definition, they cannot be seen as the counterparts of rural
migrants (Xu and Xie 2015). Substantively, as migrants move from rural to urban areas for better
life chances, their experiences of migration are perceived mainly in comparison to those from the
same origins, instead of to those of locals in destinations (Zuccotti, Ganzeboom, and Guveli
2015). In addition, existing literature has seldom compared the experiences of migrant children
and left-behind children. Indeed, the costs and benefits of bringing children to cities or leaving
them behind in the countryside are evaluated by millions of rural families every year.
Second, even those who adopted the origin-destination approach in their group comparison
relied mostly on census data, with limited indicator (i.e., school enrollment) of children's
wellbeing (Liang and Chen 2007; Wu and Zhang 2015). Existing evidence has pointed to the
China’s Internal Migration and the Wellbeing of Rural-origin Children
4
complicated and multifaceted impacts of migration on a wide variety of child outcomes. For
instance, Wu and Zhang (2015) found that rural migrant children are even less likely to be
enrolled in schools than left-behind children, which begs the question of why migrant parents
would bring along their children to cities in the first place. Holding a more optimistic view, Xu
and Xie (2015) claimed positive effects of migration on children’s objective wellbeing and no
negative effects on their subjective wellbeing. However, what they cannot explain is why most
migrant parents would still choose to leave their children behind in the countryside. These
seemly contradictory findings indicate the inappropriateness of focusing on particular outcomes,
which may lead to oversimplified interpretations.
Last but not the least, although increasingly more studies have been devoted to identifying
the signs and sizes of the so-called migration impacts, far less attention has been paid to the
context and process through which these impacts are generated. China’s internal migration
process is closely linked to the specific socioeconomic and institutional context of the country.
On one hand, the large rural-urban inequalities in resource distribution have motivated migrant
parents to bring their children to cities for better education and health care; on the other hand, the
hukou system and associated social exclusions in cities have forced the majority to leave their
children behind in the countryside (Chan and Buckingham 2008; Chan and Zhang 1999; Wu and
Treiman 2004; Xu and Wu 2015b). Presumably, given hukou’s crucial role in resource
distribution, we would expect rural-origin children who are able to obtain local urban hukou
status to be better off than other migrant children whose hukou status remains unchanged even
after migration. However, to our knowledge, the causal impacts of hukou conversion on
rural-origin children’s wellbeing have never been empirically examined.
Without thoughtful discussion and careful investigation of the true meaning of migration for
those involved and its underlying mechanisms, we would be looking at fragments of the story
rather than a larger picture. The present study is an attempt to remedy this problem. Based on a
comprehensive typology, we focus on the differences among four groups of rural-origin children:
migrant children with and without a local hukou living in cities, and left-behind and non-migrant
children living in the countryside. Specifically, we ask four research questions: (1) What are the
actual gains for rural children from migrating to urban areas? (2) What are the consequences for
migrant parents of leaving their children behind in the countryside? (3) more interestingly,
between the two plausible migration strategies, which one benefits rural children more? (4) And
finally, does obtaining a local urban hukou really payoff for rural-origin children? We investigate
China’s Internal Migration and the Wellbeing of Rural-origin Children
5
these questions using nationally representative data from a school-based survey, covering junior
high school students in both rural and urban China. We apply the propensity score matching
method to the data to ensure different groups of rural children are intrinsically comparable. To
reveal the benefits and costs involved in the migration process, we examine a wide gamut of
child wellbeing domains, ranging from school performance, physical and mental health, to future
aspirations. Findings in this paper bear important policy implications for related issues in China,
and can also be linked to studies on temporal migration in other countries or regions across the
world.
Literature Review and Research Questions
An important argument for migration by adults with children is the assumed better prospects of
their children after a successful migration. However, whether children actually benefit from the
migration experience remains a debatable question. China’s internal migration is ideal for
studying this issue, because in this case other factors often related to migration such as foreign
citizenship, and language and cultural differences (Dronkers and De Heus 2013) are not relevant
to a large extent. The long-existing rural-urban divide in China has encouraged some migrant
parents to bring their children with them to the big cities, where the educational resources and
health care facilities are far better than those available in their home villages. Figure 1 shows the
huge difference in educational resources offered to junior high school students in Shanghai (a
typical destination for migrants) and rural Henan (an important origin place of migrants). Xu and
Xie (2015) found that rural-to-urban migration benefited migrant children in some life domains
such as cognitive abilities and physical health. Recent studies from Western European countries
also suggested better educational attainment of low-class migrants than their counterparts in their
places of origin (Zuccotti, Ganzeboom, and Guveli 2015).
Nevertheless, moving from rural to urban areas does not mean that migrant children will
automatically benefit from the desirable infrastructure and resources in cities for two reasons.
First, social welfares and benefits are often reserved for those with local hukou status (Chan and
Buckingham 2008; Chan and Zhang 1999; Wu and Treiman 2004). Due to this institutional
restriction, rural migrant children are less likely to be enrolled in local schools (Liang and Chen
2007; Liang, Guo, and Duan 2008; Wu and Zhang 2015). Even when admitted, they are either
charged extra school fees (referred to as “sponsorship fees” zanzhu fei) or relegated to
sub-standard schools (Chan and Buckingham 2008; Lu and Zhou 2013; Xu and Wu 2015b).
China’s Internal Migration and the Wellbeing of Rural-origin Children
6
Second, as new comers, rural migrant children in cities are often treated as outsiders and
inferiors. They suffer from peer pressure and mental stress, increasing their anxiety toward future
prospects (Lu and Zhou 2013; Xiong 2015; Xu and Wu 2015b). For instance, Xiong (2015)
demonstrated that, in Shanghai, most migrant children gave up their educational aspirations after
realizing their chances of upward mobility are severely limited. Considering the complex
influences involved in the child migration process, our first research question would be:
What are the actual gains for rural children from migrating to urban areas?
30000
14
25000
12
10
20000
8
15000
6
10000
Ratio
Educational Expenditure per Student in
Juunior high schools (Yuan)
Figure 1. Educational Expenditure per Student in Junior High Schools
4
5000
2
0
0
Year
Country Avg.
Shanghai
Rural Henan
Ratio: Shanghai/Rural Henan
Data source: China Educational Finance Statistical Yearbook, 1997-2012
The empirical findings on the wellbeing of left-behind children are even more mixed. Some
studies have suggested that parent-child separation makes left-behind children extremely
vulnerable, resulting in their low educational achievements (Hu 2012; Lu 2012; Zhao, Yu, Wang,
and Glauben 2014), poor physical health and health-related behaviors (Gao, Li, Kim, Congdon,
Lau, and Griffiths 2010; Li, Liu, and Zang 2015; Wen and Lin 2012), and psychological and
developmental problems (Jia and Tian 2010; Wu, Lu, and Kang 2015). Other studies, however,
China’s Internal Migration and the Wellbeing of Rural-origin Children
7
have shown that parental migration is not necessarily harmful for children’s development (Chen,
Huang, Rozelle, Shi, and Zhang 2009; Wen, Su, Li, and Lin 2015; Xu and Xie 2015). One
possible explanation is that these children continue to be taken care of by their grandparents or
other caregivers. Moreover, migrant parents would send remittances back home, so left-behind
children would tend to have more material and financial resources to overcome constraints and
increase their human capital investments than other rural children whose parents live in the
countryside with them (Hu 2012; Mu and Brauw 2015). Since the vast majority of children in
rural China are currently left behind by migrant parents, their livelihood and psychological states
have caused widespread concern among the public. Therefore, our second research question is:
What are the consequences for migrant parents of leaving their children behind
in the countryside?
Previous literature has largely neglected the fact that the migration process involves two
steps for rural families with children. Rural parents would first have to decide whether or not to
migrate, and then they must decide whether to bring their children along or leave them behind.
Although the latter decision is made by millions of migrant parents in China, few studies have
attempted to investigate its effect on the wellbeing of migrant and left-behind children. Liang and
Chen (2007), Wu and Zhang (2015), the two exceptions, both focus on school enrollment using
census data. They show a significantly higher risk of dropping out for migrant children who went
along with their parents to cities. This may partially explain some migrant parents’ decision to
leave their children behind in the countryside despite the pain and problems of family separation.
For the first time, based on school-based data, we will adopt the origin-destination approach to
investigate the third question:
Between the two plausible migration strategies, which one benefits rural children
more?
A distinctive feature of China’s internal migration is that, the hukou system has been acting
constantly as an invisible barrier hindering migrants’ assimilation process, and forcing migrant
parents to leave their children behind in the countryside. For example, entrance examinations of
high schools and colleges must be taken in the locality of hukou registration and not in the
China’s Internal Migration and the Wellbeing of Rural-origin Children
8
locality of residence, which means that migrant children must return to their hometown if they
wish to pursue further studies. Therefore, the institutional barriers imposed by the hukou system
are key to the migration strategies adopted by migrant families and their consequences for rural
children in China. It is worth mentioning that, though rare, rural-origin children can actually
obtain an urban hukou. Their hukou is converted once their parents meet certain criteria, 1 or
when cities expand and their villages are incorporated into urban areas (Zhang and Treiman
2013). To the extent that the lack of a local hukou indeed puts rural migrant children at a
disadvantage, we should observe better outcomes for those rural-origin migrant children who are
able to convert their hukou status and gain local citizenship than for those who fail to do so. This
speculation leads us to the last research question:
Does obtaining a local urban hukou really pay off for rural-origin children?
Analytical Framework and Hypotheses
To answer the research questions raised in the previous section, it is necessary to develop a
comprehensive typology of all the rural-origin children, as well as appropriate comparison
strategies for causal assessment. A complete rural-to-urban migration in China generally consists
of three processes: parental migration, child migration, and finally, hukou conversion. Figure 2
shows these three migration processes and also the resulting typology of rural-origin children.
Note that these three processes follow a sequential order, rather than simply a temporal order. For
instance, the decisions of parental migration and child migration may be made at the same time.
As clearly suggested by the analytical framework, internal migration in China is a complicated
process consisting of several phrases, so lumping different types of migration effects together
will severely contaminate our empirical analysis and limit our understanding of their
consequences.
Because migrant parents may either bring their children with them to cities or leave them
behind in the countryside, we are actually dealing with two different types of migration effects.
1
Mainly through attaining tertiary or technical secondary education, or joining the People’s Liberation
Army (Wu and Treiman 2004).
China’s Internal Migration and the Wellbeing of Rural-origin Children
9
Figure 2. Migration Processes and Typology of Rural-origin Children
Parental Migration
Yes
Child Migration
Yes
Local Hukou Attainment
Yes
Converters
No
No
No
Migrants
Left-behinds
Non-migrants
One is the effect of parental migration, and the other is the effect of child migration (or family
migration, since children can only migrate with their parents). Another type of migration process
that is consistently neglected by scholars is the attainment of a local hukou, which is the final
step of a successful migration. The value of migration is expected to be the largest for the group
of rural-origin children who are able to gain a local hukou. It is therefore necessary to distinguish
among these different types of migration patterns, and be specific about the group we are
interested in and the reference group under different circumstances.
Much of the migration/immigration study has been centered on how migrants/immigrants
have fared compared with local natives, and measuring their “success” as the extent to which
they have closed the socioeconomic gap with natives, and assimilated into destination societies.
However, as criticized by Zuccotti, Ganzeboom, and Guveli (2015), “…this may not be the
perspective that migrants themselves find most relevant. People do not move to compete with
other groups in the destination society but to improve their life chances—and their
children’s—relative to what they would have been in the origin society.” In addition, Xu and Xie
China’s Internal Migration and the Wellbeing of Rural-origin Children
10
(2015) also pointed out that children from non-migrant families in rural areas are a more suitable
reference group for migrant children as they are “potentially exposable” (Holland 1986) to
migration experience. Therefore, to assess and understand the causal impacts of rural-to-urban
migration on children’s wellbeing, it is necessary to examine the benefits (or costs) for migrant
children in destination cities compared to their non-migrant counterparts in the countryside or the
left-behind children of migrated parents (Feliciano 2005; Zuccotti, Ganzeboom, and Guveli
2015).
To sum up, for the purpose of examining the effect of parental migration on rural children,
we contrast left-behind children with non-migrant children who both live in rural areas; to
examine the effect of their own migration on rural children, we contrast migrant children in cities
with non-migrant children in the countryside; to compare the two migration strategies of
bringing their children along with them to cities and leaving them behind in the countryside, we
contrast left-behind children with migrant children, using the latter as the reference group; finally,
to assess the value of obtaining a local urban hukou, we contrast those rural-origin children who
currently hold a local (urban) hukou with their peers who do not.
Based on these comparison strategies for causal inferences, we propose several research
hypotheses. Our focus on child developmental outcomes can be broadly divided into four
domains: school performance, physical health, mental health, and future aspirations. Notably,
one of the major contributions of this paper is that it attempts to uncover the impacts of
migration on children’s future aspirations. Studies have shown that higher aspirations often
provide migrants with stronger motivations for upward mobility, and can partly explain their
academic success despite their disadvantaged family socioeconomic background (Xu and Wu
2015a). If the destination society is more accepting of newcomers, migrants and their children
would be more confident and would have higher hopes for their future. In turn, if migrant
families experience segregation and discrimination in cities, they would exhibit a lower level of
aspiration and feel more anxious toward their future. Therefore, rural children’s future
aspirations are arguably a direct reflection of their subjective feelings and attitudes toward
migration, whether it involves their parents only or involves them as well.
China’s Internal Migration and the Wellbeing of Rural-origin Children
11
Better educational resources and health care facilities in destination cities should benefit
migrant children. But facing an unfamiliar environment and being subjected to institutional
segregation and discrimination could generate extra psychological burdens on migrant children,
even affecting their perceived life chances for upward mobility. Thus, we expect:
H1: Compared to rural non-migrant children, migrant children in cities would enjoy some
advantages in school performance and physical health, but also face some disadvantages in
mental health and future aspirations.
Parent-child separation could result in a lack of love and care and parental supervision, and
hence could be harmful for children’s schooling and health. However, the remittances sent back
by migrant parents could significantly increase family income and children’s education
opportunities. Therefore, we would expect disruptive effects of parental migration on left-behind
children’s physical and mental health, but mixed results for their school performance. Also, as
these left-behind children are not directly exposed to segregation and discrimination in cities as
their migrant peers are, they may not necessarily have lower future aspirations. Thus, we expect:
H2: Compared to their non-migrant counterparts, left-behind children in rural areas would enjoy
some but limited advantages in school performance and future aspirations, but they would be
disadvantaged in physical and mental health.
It is just as difficult for a migrant from rural China to obtain a local urban hukou as it is for
an illegal migrant in the US to obtain a green card. With a local urban hukou, we would expect
the children of “successful rural migrants” to outperform rural migrant children in cities without
a local urban hukou in terms of school performance, and to also have higher future aspirations,
because the former face far less institutional boundaries. However they are still subject to all the
stresses and strains of the migration process.
H3: Compared to rural migrant children without a local hukou, rural-origin children who
successfully obtain a local hukou would enjoy advantages in school performance and future
aspirations, but the two groups of children would not exhibit any differences in terms of physical
and mental health.
As for the comparison between migrant children and left-behind children, we do not lay out
any prior assumptions because of a lack of theoretical conjectures and empirical evidence in the
previous literature. Table 1 clearly demonstrates the hypothetical benefits and costs that
parental/child migration and hukou conversion bring to different aspects of child wellbeing.
China’s Internal Migration and the Wellbeing of Rural-origin Children
12
Table 1. Hypothetical Benefits and Costs of the Migration Process on Selected Child
Development Domains
Child migration
(Migrants vs.
Non-migrants)
Parental Migration
(Left-behinds vs.
Non-migrants)
Local Hukou Attainment
(Converters vs. Migrants)
+
+
-
+/+/-
+
+/+/+
School performance
Physical health
Mental health
Future aspirations
Data, Variables, and Methods
Data
The empirical analysis is based on data from the first wave of the 2014 China Education Panel
Survey (CEPS), 2 a nationally representative school-based survey on junior high school students
in China. Employing a multi-stage stratified PPS sampling design, the survey first selected 28
counties/districts across the country, and then sampled four junior high schools within the
geographic boundaries of each county/district. For each school, four classes (two 7th grade
classes and two 9th grade classes) were sampled, and all of the students within each class would
enter the survey. Overall, the survey covered approximately 20,000 students in 112 schools from
both urban and rural regions, enabling us to locate rural students in both origins and destinations.
Moreover, designed deliberately to focus on migrant children, the survey also oversampled those
counties/districts with large in-migrant populations. Particularly, we would like to highlight the
fact that this was a school-based survey, which suggests only children who were officially
enrolled in schools were sampled. Considering that rural migrant children in cities had a higher
dropout rate (Liang and Chen 2007; Wu and Zhang 2015), this group tended to be positively
selected in our sample. We are aware of this data limitation, and will discuss the implications for
our findings in the conclusion.
In this paper, our primary interest is the impact of parental or child migration, as well as
hukou conversion, on rural children’s/students’ wellbeing outcomes. Hence, our analytic sample
consists of four groups of rural-origin children: children who migrate with their parents and
2
http://chinaeps.org/index.php?r=index/index
China’s Internal Migration and the Wellbeing of Rural-origin Children
13
successfully acquire a local hukou; children who migrate with their parents but fail to acquire a
local hukou; children who are left behind by migrant parents; and non-migrant children who are
not exposed to any direct influence of migration. Informed by previous research, we define
rural-to-urban migrant children (N=1,769) as those who currently live in urban areas while
holding a non-local rural hukou (excluding those moving across district boundaries within the
city); we define hukou converters (N=150) as those who were born with a non-local rural hukou
but currently hold a local urban hukou. Both left-behind children (N=1,246) and non-migrant
children (N=3,549) are local rural hukou holders in rural areas. Left-behind children do not live
with their parents, and at least one of their parents is engaged in non-farm work. 3 Any other
groups that do not fall into one of these four categories are excluded from our sample.
Variables
Considering the complex influences of migration on children’s wellbeing, we examine a variety
of child developmental outcomes, rather than focusing on one particular aspect. School
performance is one of the most important domains of child development. We adopt three related
measures for this outcome. First, cognitive abilities are measured by standardized scores from a
15-minute in-class assessment on students’ learned reasoning abilities in three areas most linked
to academic success in school: verbal, numerical, and graphical. Second, school engagement is
measured by the number of hours that students spend on homework on an average day (including
weekdays and weekends). Third, school attachment is measured by scores derived from a
12-item scale asking about students’ feelings and experiences of school life. Health is another
essential domain of child development. We measure physical health based on a five-point Likert
scale of students’ health status, 4 and also measure mental health using the standard five-item
depression scale widely used by other scholars. In addition, we also adopt two measures to
capture the influences of migration on rural children’s perceived life chances: educational
aspirations and confidence about the future.
3
We adopt this alternative measure, as we do not have information on parents’ current place of
residence.
4
The values of this variable are based on responses made by the parents (or any other caregivers who
completed the questionnaire for parents). If missing, we use students’ own responses instead.
China’s Internal Migration and the Wellbeing of Rural-origin Children
14
We also consider a wide range of individual characteristics and family background as
control variables. The individual characteristics include four dummy variables: students’ sex
(male=1), grade level (9th grade=1), ethnicity (minority=1), and prior achievement (ever repeated
a grade during primary school=1). In addition, age is coded into five categories (12, 13, 14, 15,
16). Furthermore, the model also includes several variables on family background. Place of
origin is coded into four categories (east, northeast, central, west), parental education is coded
into three categories (primary or lower, junior high, senior high or above), number of siblings is
coded into four categories (0, 1, 2, 3 or above), family economic status before primary school is
a dummy variable (poor=1), 2 family cultural capital is measured by the number of books at
home, which is a categorical variable (very few, few, moderate, many, a lot).
Table 2 shows the descriptive statistics on the aforementioned variables by rural families’
migration status. As shown in the table, hukou converters outperform the other three groups of
rural-origin children on most of the wellbeing domains, especially cognitive abilities. However,
their family background is also significantly better than that of other groups, so it is not clear
whether or not hukou conversion has a net advantage after controlling for other factors. Apart
from this highly selected group, rural migrant children seem to be better off than their
counterparts in origins, particularly in terms of school performance and physical health. There
seems to be no systematic differences between left-behind children and non-migrant children in
rural areas in terms of both children’s outcomes and family background. Nevertheless, we do
find that left-behind children have the worst physical and mental health among all four groups of
children.
Methods
Any migration decision is selective in essence (Borjas 1987). The four groups of rural-origin
children are not directly comparable, because their developmental outcomes can be
simultaneously affected by migration status and other individual or family characteristics. For
instance, rural families with better economic resources or social capital are more likely to
overcome institutional obstacles and successfully send their children to high-quality urban
China’s Internal Migration and the Wellbeing of Rural-origin Children
15
Table 2. Descriptive Statistics on Selected Variables, by Migration Status
Converters
Outcomes
Cognitive abilities
Hours/day spent on
homework
Attachment to school
Health
Depression
Educational aspirations
Confidence about future
Individual characteristics
Male
Age
Grade (9th=1)
Minority (yes=1)
Grade retention (yes=1)
Family background
Place of origin
East
Northeast
Central
West
Parental education
Primary or lower
Junior high
Senior high and above
No. of siblings
0
1
2
3 and above
Poor family (yes=1)
No. of Books at home
Very few
Few
Moderate
Many
A lot
N
Migrants
Left-behinds
Non-migrants
0.293(0.731)
3.030(1.299)
-0.023(0.852)
2.585(1.446)
-0.250(0.728)
2.542(1.711)
-0.264(0.751)
2.458(1.682)
2.164(0.326)
4.287(0.789)
2.038(0.903)
5.573(1.961)
3.306(0.727)
2.169(0.347)
4.216(0.860)
2.085(0.831)
5.414(2.033)
3.163(0.712)
2.120(0.355)
3.902(0.872)
2.204 (0.767)
5.421(2.084)
3.099(0.729)
2.158(0.348)
3.962(0.906)
2.143(0.699)
5.365(2.055)
3.163(0.726)
0.520(0.501)
13.75(1.158)
0.493(0.502)
0.040(0.197)
0.047(0.212)
0.509(0.500)
13.92(1.226)
0.397(0.489)
0.054(0.225)
0.204(0.403)
0.549(0.498)
13.97(1.288)
0.473(0.499)
0.129(0.336)
0.364(0.481)
0.512(0.500)
14.14(1.272)
0.526(0.499)
0.133(0.340)
0.291(0.454)
0.680(0.468)
0.033(0.180)
0.000(0.000)
0.287(0.454)
0.369 (0.483)
0.057(0.231)
0.271(0.445)
0.303(0.460)
0.005(0.069)
0.026(0.161)
0.544(0.498)
0.425(0.494)
0.001(0.034)
0.102(0.303)
0.483(0.500)
0.413(0.493)
0.013(0.115)
0.320(0.468)
0.667(0.473)
0.125(0.331)
0.558(0.497)
0.317(0.465)
0.160(0.366)
0.588(0.492)
0.252(0.434)
0.168(0.374)
0.613(0.487)
0.219(0.413)
0.533(0.501)
0.380(0.487)
0.080(0.272)
0.007(0.082)
0.233(0.424)
0.276(0.447)
0.523(0.500)
0.155(0.362)
0.046(0.209)
0.289(0.453)
0.165(0.371)
0.571(0.495)
0.220(0.414)
0.044(0.205)
0.460(0.499)
0.192(0.394)
0.592(0.491)
0.170(0.375)
0.046(0.210)
0.499(0.500)
0.013(0.115)
0.073(0.262)
0.387(0.489)
0.313(0.465)
0.213(0.411)
0.105(0.306)
0.134(0.341)
0.395(0.489)
0.239(0.426)
0.128(0.335)
0.247 (0.432)
0.242(0.429)
0.356(0.479)
0.128(0.334)
0.027(0.163)
0.241(0.428)
0.224(0.417)
0.380(0.485)
0.116(0.320)
0.039(0.195)
150
1769
1246
3549
Note: Standard deviations are in parentheses.
China’s Internal Migration and the Wellbeing of Rural-origin Children
16
schools, therefore it is not clear whether better educational outcomes of migrant children are due
to the migration per se or family resources. Reverse causality could also be a problem,
considering that some migrant parents may strategically choose to bring the smarter (or healthier)
children with them to cities for better educational opportunities, and leave the not-so-smart (or
not-so-healthy) ones behind to reduce costs. To address these endogeneity concerns, scholars
would adopt either the instrumental variable approach (Hu 2012; Zhao, Yu, Wang, and Glauben
2014) or the propensity score matching approach (Chen et al. 2009; Xu and Xie 2015). In this
paper, we use the latter. 5
All of the individual and family characteristics we mentioned in the previous section are
incorporated into the propensity score matching analysis as matching covariates. After estimating
the propensity scores, we proceed to match the treatment group and the control group.
Particularly, we choose the optimal (full) matching method for its two desirable features over
other matching methods.
First, most matching algorithms involve a series of small decisions, and these decisions are
made one at a time without reconsidering earlier decisions (Rosenbaum 2002). This process
makes final results sensitive to arbitrary decisions such as choosing caliper sizes, or deciding
bandwidth values and trimming schedules; it also makes the quality of the matches sensitive to
the order in which treated subjects are matched. By applying network flow theory, optimal
matching can avoid this issue by taking into account the overall set of matches when choosing
individual matches. In other words, in the optimal matching process, later decisions can alter
earlier ones. Specifically, optimal matching aims to develop matched sets in such a way that the
matching can optimize or minimize the total distance of propensity scores (Guo and Fraser 2014).
For optimal full matching 6, each treated case is matched to one or more controls, and similarly,
each control case is matched to one or more treated subjects.
5
We also conduct naïve OLS regression analysis, and results are reported in the Appendix.
The other two types of optimal matching methods are pair matching and variable matching. However,
they are generally not optimal when compared to full matching using the same data (Rosenbaum 2002).
Nonetheless we also tried pair matching and variable matching. The imbalance check results show that
full matching performed the best among the three.
6
China’s Internal Migration and the Wellbeing of Rural-origin Children
17
Second, the use of most matching algorithms requires considerable overlapping of estimated
propensity scores for treatment and control groups. Violation of this overlap assumption is a
major source of evaluation bias (Heckman, Ichimura, and Todd 1997). However, optimal
matching does not require a sizable common support region to work (Guo and Fraser 2014).
Following the well-established implementation steps suggested by scholars (Caliendo and
Kopeinig 2008), once the propensity scores are estimated, we first show the density distribution
of propensity scores for both treatment and control group to check the overlap between the two.
As can be seen in the Appendix, it is likely that the problem of a narrow common support region
under some circumstances would produce a nontrivial loss of matched subjects if we use greedy
matching or kernel-based matching. This justifies our use of the optimal matching method, as we
are not made to drop subjects falling outside the range of the common support region, or to use a
trimming strategy to discard sparse subjects with extreme propensity scores.
How do we evaluate the effectiveness of such matching methods? According to Haviland,
Nagin, and Rosenbaum (2007), the level of bias reduction can be calculated by comparing the
absolute standardized differences in covariate means before and after optimal matching. The
imbalance on covariate X is measured by 𝑑𝑑𝑥𝑥
The formulas are
𝑑𝑑𝑥𝑥 =
before matching and by 𝑑𝑑𝑥𝑥𝑚𝑚 after matching.
�𝑀𝑀𝑥𝑥𝑥𝑥 − 𝑀𝑀𝑥𝑥𝑥𝑥 �
2
2 )⁄2
+ 𝑆𝑆𝑥𝑥𝑥𝑥
�(𝑆𝑆𝑥𝑥𝑥𝑥
𝑑𝑑𝑥𝑥𝑥𝑥 =
|𝑀𝑀𝑥𝑥𝑥𝑥 − 𝑀𝑀𝑥𝑥𝑥𝑥 |
2
2 )⁄2
+ 𝑆𝑆𝑥𝑥𝑥𝑥
�(𝑆𝑆𝑥𝑥𝑥𝑥
where subscripts t, p, and c denote the treatment group, the potential control group, and the
control group, respectively. So 𝑀𝑀𝑥𝑥𝑥𝑥 and 𝑀𝑀𝑥𝑥𝑥𝑥 are the means of covariate X for the treatment
group and the potential control group, and 𝑀𝑀𝑥𝑥𝑥𝑥 is the unweighted mean of means of X for the
controls matched to treated cases. The denominator is the overall standard deviation of the mean
values. 𝑑𝑑𝑥𝑥 and 𝑑𝑑𝑥𝑥𝑥𝑥 can be interpreted as the differences in standard deviation of X between
the treatment and control groups, and thus are directly comparable (Guo and Fraser 2014).
China’s Internal Migration and the Wellbeing of Rural-origin Children
18
Results
Matching
We first compare migrant children with non-migrant children to examine the effects of child
migration (with migrant children being the treatment group), before comparing left-behind
children with non-migrant children to examine the effects of parental migration (with left-behind
children being the treatment group). Then, we compare left-behind children with migrant
children to examine the consequences of leaving children behind (with left-behind children being
the treatment group). Finally, we compare hukou converters with other rural-migrant children
without a local hukou to examine the benefits of hukou conversion (with hukou converters being
the treatment group). When comparing left-behind children with migrant children, we are
actually speculating about how a left-behind child would have performed had he or she been
brought to cities with their migrant parents (i.e., not receiving the treatment).
Table 3 reports the sample description and bivariate test on variables for four comparison
scenarios, and Table 4 presents the logistic regressions predicting the propensity scores.
Although many interesting results are presented in this table, we cannot go through each of the
comparisons due to limited space. Instead, we take the comparison of two migration strategies
(i.e., leaving children behind or bringing them to cities) as an example. Specifically, we find that
boys and girls are equally likely to be left behind in the countryside. That is to say, there is no
evidence of son preference in migrant parents’ arrangement for their children. 7 Other things
being equal, older or higher grade rural children are more likely to be left behind. This result
makes sense considering the fact that, although cities are responsible for providing nine-year
compulsory education for migrant children (as required by state policy introduced in 2001), 8
they have no obligations to admit them into local senior high schools.
7
However, since only enrolled students were sampled in our survey and if more migrant boys dropped
out of urban schools than migrant girls (because they have a higher chance of earning money in
unskilled work than migrant girls), it can also explain the insignificant gender difference we observed.
8
“Guowuyuan Guanyu Jichu Jiaoyu Gaige yu Fazhan de Jueding” (“State Council’s Decisions on Reforms
and Development of Basic Education”), 29 May, 2001.
China’s Internal Migration and the Wellbeing of Rural-origin Children
19
Table 3. Sample Description and Bivariate Test of Variables
Migrants vs.
Left-behinds vs.
Left-behinds vs.
Hukou Converters vs.
Non-migrants
Non-migrants
Migrants
Migrants
% of
treated
Gender
Bivariate
𝜒𝜒 2
test
% of
treated
0.871
Bivariate
𝜒𝜒 2
test
% of
treated
0.023
Bivariate
𝜒𝜒 2
test
% of
Bivariate
treated
𝜒𝜒 2 test
0.032
0.802
Female
33.37
24.49
39.30
7.66
Male
33.16
27.36
43.15
7.97
Age
0.000
0.002
0.000
0.013
12
30.24
31.47
51.44
9.71
13
44.28
26.94
31.70
8.38
14
29.04
26.13
46.37
5.96
15
31.26
24.30
41.38
9.93
16
26.31
22.63
45.04
3.14
Grade
0.000
0.001
0.000
0.021
th
38.81
28.09
38.11
6.65
th
27.33
23.98
45.62
9.54
7 grade
9 grade
Minority
0.000
0.734
0.000
0.471
No
35.23
26.07
39.33
7.92
Yes
16.75
25.43
62.89
5.94
Grade retention
0.000
0.000
0.000
0.000
No
35.88
23.96
36.03
9.22
Yes
25.90
30.48
55.65
1.90
Place of origin
0.000
0.000
0.000
0.000
East
99.39
60.00
0.91
13.51
Northeast
21.60
8.33
24.81
4.76
Central
21.87
28.33
58.55
0.00
West
26.76
26.50
49.67
7.43
Parental
0.000
0.053
0.000
0.000
education
Primary or
27.07
24.97
47.27
0.89
Junior high
31.21
25.21
42.62
4.64
Senior high and
41.92
28.81
35.93
15.15
lower
above
No. of siblings
0.000
0.001
0.000
0.000
0
41.79
23.14
29.54
14.06
1
30.56
25.30
43.49
5.80
2
31.28
31.28
50.00
4.20
3 and above
33.06
25.11
40.44
1.22
Poor family
0.000
0.017
0.000
0.148
China’s Internal Migration and the Wellbeing of Rural-origin Children
20
No
41.44
27.46
34.85
8.38
Yes
22.39
24.45
52.86
6.41
No. of Books at
0.000
0.097
0.000
0.000
home
N
Very few
17.81
26.51
62.47
1.07
Few
22.94
27.50
56.03
4.44
Moderate
34.10
24.72
38.83
7.67
Many
50.72
27.94
27.37
10.02
A lot
61.85
19.54
13.03
12.36
5,318
4,795
3,015
1,439
Children with poor prior academic performance (grade repeaters) are less likely to be
brought to cities, suggesting rural parents do make strategic migration decisions based on their
children’s educational and earning potential. Minorities and those originating from central or
western regions are far more likely to be left behind, probably due to the longer traveling
distance to destination cities. Surprisingly, parental education has no net influence on children’s
migration status, 9 while a family’s cultural capital (number of books at home) significantly
increases the probability of child migration. Children with more siblings are usually left behind,
probably because migrants with more children are less likely to bring all of their children with
them because of the high living costs in cities. A family’s economic condition is also an essential
factor. Poor families tend to leave their children at home as they cannot afford the extra costs of
bringing them to cities.
As Table 5 shows, the covariate imbalances are greatly reduced for almost all the variables,
suggesting that optimal full matching worked reasonably well, and efficiently balanced the
treated samples and control samples. Taking the poor family indicator in the comparison between
left-behinds and migrants as an example, the treatment and control groups differ on this variable
by more than 38 percent of a standard deviation before matching, whereas the standard bias is
reduced to only 0.9 percent after full matching.
9
However, we do observe a significant positive effect of parental educational attainment on hukou
conversion. This is consistent with the common wisdom that tertiary education is probably one of the
most important pathways for acquiring a local urban hukou.
China’s Internal Migration and the Wellbeing of Rural-origin Children
21
Table 4. Logistic Regression Models Predicting Propensity Scores
Male
Age (ref: 12)
13
14
15
16
Grade (9th grade=1)
Minority
Grade retention (yes=1)
Place of origin (ref: east)
Northeast
Central
West
Parental education (ref:
primary or lower)
Junior high
Senior high and above
No. of siblings (ref: 0)
1
2
3 and above
Poor family (yes=1)
No. of Books at home (ref:
very few)
Few
Moderate
Many
A lot
Constant
N
Migrants vs.
Non-migrants
Left-behinds vs.
Non-migrants
Left-behinds vs.
Migrants
Hukou
Converters vs.
Migrants
0.908
1.181*
1.180
0.962
2.569***
0.727*
0.296***
0.926
2.927***
0.626**
0.225***
0.226***
4.969***
0.523*
0.119***
0.209**
5.684***
0.424*
0.073***
0.099***
0.279***
1.174
4.113***
6.521***
0.313*
1.100
4.229*
0.954
0.683
1.488**
2.362***
0.459
0.001***
0.075***
59.898***
0.264**
0.002***
0.298*
148.404***
-
0.002***
0.320
103.572***
0.471*
0.926
1.069
1.112
4.207*
1.194
1.194
1.048
12.316***
0.605**
0.917
1.468*
0.465***
0.731
1.266
1.889**
0.460*
0.587
0.934
1.389
0.117*
0.541***
0.791**
1.610***
1.104
1.304
1.078
0.773
3.602
2.018**
0.931
0.480**
5.012*
3.951***
1.144
0.293***
5.148*
6.253***
0.718
0.108***
5.552*
108.663***
1.516
0.016***
0.009***
5,318
4,795
3,015
1,439
Note: Coefficients are odds ratios; robust standard errors adjusted for the clustering effect on schools are not
reported; *** p<0.001, ** p<0.01, * p<0.05.
China’s Internal Migration and the Wellbeing of Rural-origin Children
22
Table 5. Covariate Imbalance Check between Treatment and Control Groups, before and after Optimal Full Matching
Migrants vs. Non-migrants
Male
Age
12
13
14
15
16
Grade
Minority
Grade retention
Place of origin
East
Northeast
Central
West
Parental education
Primary or lower
Junior high
Senior high and
above
No. of siblings
0
1
Left-behinds vs.
Non-migrants
Left-behinds vs. Migrants
Hukou Converters vs.
Migrants
𝑑𝑑𝑥𝑥
Before
matching
𝑑𝑑𝑥𝑥𝑥𝑥
After
matching
𝑑𝑑𝑥𝑥
Before
matching
𝑑𝑑𝑥𝑥𝑥𝑥
After
matching
𝑑𝑑𝑥𝑥
Before
matching
𝑑𝑑𝑥𝑥𝑥𝑥
After
matching
𝑑𝑑𝑥𝑥
Before
matching
𝑑𝑑𝑥𝑥𝑥𝑥
After
matching
0.005
0.009
0.075
0.024
0.079
0.015
0.021
0.053
0.050
0.290
0.104
0.052
0.137
0.261
0.275
0.203
0.003
0.006
0.000
0.007
0.003
0.008
0.008
0.019
0.108
0.026
0.004
0.050
0.078
0.107
0.011
0.155
0.020
0.013
0.014
0.017
0.002
0.005
0.045
0.005
0.157
0.263
0.108
0.001
0.059
0.153
0.264
0.359
0.015
0.010
0.030
0.003
0.031
0.041
0.029
0.026
0.087
0.057
0.131
0.157
0.274
0.195
0.065
0.489
0.050
0.063
0.027
0.007
0.025
0.012
0.016
0.011
1.076
0.170
0.448
0.232
0.003
0.012
0.064
0.055
0.068
0.313
0.122
0.023
0.032
0.006
0.008
0.002
1.057
0.151
0.578
0.254
0.001
0.028
0.024
0.035
0.654
0.112
0.863
0.036
0.067
0.024
0.015
0.048
0.122
0.112
0.222
0.014
0.001
0.013
0.024
0.050
0.079
0.007
0.019
0.016
0.098
0.061
0.143
0.009
0.014
0.023
0.452
0.493
0.746
0.001
0.020
0.021
0.201
0.140
0.004
0.036
0.072
0.042
0.004
0.008
0.272
0.098
0.003
0.029
0.541
0.290
0.032
0.030
23
China’s Internal Migration and the Wellbeing of Rural-origin Children
2
3 and above
Poor family
No. of Books at home
Very few
Few
Moderate
Many
A lot
0.040
0.002
0.440
0.027
0.030
0.025
0.127
0.010
0.078
0.010
0.030
0.003
0.167
0.008
0.359
0.029
0.020
0.009
0.234
0.246
0.126
0.000
0.000
0.052
0.366
0.237
0.030
0.327
0.325
0.008
0.013
0.005
0.006
0.047
0.015
0.043
0.051
0.037
0.068
0.004
0.018
0.030
0.073
0.000
0.381
0.280
0.081
0.290
0.384
0.023
0.013
0.009
0.002
0.006
0.395
0.200
0.016
0.168
0.227
0.006
0.004
0.014
0.116
0.127
Note: 𝑑𝑑𝑥𝑥 and 𝑑𝑑𝑥𝑥𝑥𝑥 are absolute standardized differences in covariate means between the treatment group and the control group, 𝑑𝑑𝑥𝑥 for use before
matching and 𝑑𝑑𝑥𝑥𝑥𝑥 for use after matching.
Table 6. Estimated Treatment Effects on Various Outcomes after Optimal Full Matching
Migrants vs. Non-migrants
ATE
P-value
Left-behinds vs.
Non-migrants
ATE
P-value
Left-behinds vs. Migrants
ATE
P-value
Hukou Converters vs.
Migrants
ATE
P-value
Cognitive abilities
0.287
0.000
0.036
0.145
-0.276
0.000
0.144
0.037
Hours/day spent on
0.090
0.000
0.084
0.012
-0.025
0.027
0.166
0.119
homework
Attachment to school
-0.053
0.091
-0.096
0.011
-0.093
0.027
-0.001
0.442
Health
0.218
0.000
-0.063
0.042
-0.317
0.000
0.084
0.320
Depression
-0.036
0.346
0.070
0.039
0.050
0.103
-0.136
0.064
Educational aspirations
-0.073
0.037
0.040
0.117
0.061
0.060
0.077
0.196
Confidence about future
-0.171
0.000
-0.098
0.006
-0.021
0.247
0.036
0.187
Note: The P-values are derived from the Hodges-Lehmann aligned rank tests (one-tailed). ATEs are standardized mean differences between the treatment
group and the control group after matching.
China’s Internal Migration and the Wellbeing of Rural-origin Children
24
Causal Impacts of Migration on Child Wellbeing
When X is the outcome variable of interest, 𝑑𝑑𝑥𝑥𝑥𝑥 is also recognized as the average treatment
effect (ATE) after optimal matching. The significance of the ATE is revealed by performing the
Hodges-Lehmann aligned rank test (Lehmann and D'Abrera 2006). Table 6 summarizes the
estimated treatment effects on child developmental outcomes after optimal full matching.
1. Do Rural Children Benefit from Migration to Cities?
As shown in Table 6, moving from the countryside to cities does generate some beneficial
outcomes for rural children as we would expect. Their cognitive abilities test scores are higher
than those of non-migrants by close to 30 percent of a standard deviation; their time spent on
homework is longer by almost 10 percent of a standard deviation; and their health is better by 22
percent of a standard deviation. But these benefits also come with a price. Being treated as
outsiders and inferiors, migrant children in cities often face major challenges in the local
education system, such as admission discrimination (Goodburn 2009; Wu and Zhang 2015),
counter-school culture (Xiong 2015), and loneliness (Lu and Zhou 2013). A study relying on the
same data (Xu and Wu 2015b) showed that hukou-based school segregation has largely driven
migrant children to sub-standard schools and exposed them to various negative peer influences.
Our results here further confirm that these factors could bring adverse impacts on migrant
children’s future aspirations, once they realize they have few opportunities for upward mobility.
Their educational aspirations are lower than those of non-migrants by more than 7 percent of a
standard deviation, and their confidence about the future is lower by approximately 17 percent of
a standard deviation. These findings are in line with our first research hypothesis, and support the
classic assumption that migrants are motivated to achieve a better life for themselves and their
children.
2. Does being Left Behind in the Countryside Hurt Children’s Wellbeing?
With regard to the effects of parental migration on the wellbeing of those left behinds, previous
literature has argued for both the negative effects resulting from parent-child separation (Gao et
China’s Internal Migration and the Wellbeing of Rural-origin Children
25
al. 2010; Wu, Lu, and Kang 2015) and the positive effects resulting from remittances (Hu 2012;
Mu and Brauw 2015). However, based on empirical evidence, money cannot compensate for
parental affection. As Table 6 shows, contrary to people’s expectation, the remittances sent back
by migrant parents do not improve rural children’s school performance by much. Although
left-behind children spend slightly more time on homework than their non-migrant peers (by
approximately 8 percent of a standard deviation), they do not show significantly higher cognitive
abilities. Even worse, they tend to exhibit weaker attachment to schools (by close to 10 percent
of a standard deviation), which is not a good sign, as weak school attachment is often closely
related to higher rates of dropping out (South, Haynie, and Bose 2007). These results suggest that
even if remittances could bring some benefits for rural children, it is likely that these positive
effects are more or less offset by the adverse effects of family separation. For instance, among
left-behind children, the level of depression is significantly higher by 7 percent of a standard
deviation, and the physical health is also worse by approximately 6 percent of a standard
deviation. Interestingly, even though left-behind children are not directly exposed to migration
experience themselves, like migrant children, they also show lower confidence about the future
albeit to a less extent (by approximately 10 percent of a standard deviation). This finding is
somewhat ironic, as migration is initially motivated by people’s desire to find better life chances
yet somehow it ends up smashing their dreams and increasing their anxiety toward the future.
Therefore, our second research hypothesis is only partially supported.
3.To Go or Not to Go?
Millions of Chinese migrant parents face the dilemma of whether to bring their children with
them to cities or to leave them behind in the countryside. Which of the two migration strategies
would benefit their children more? Results in Table 6 reveal that it is not a simple zero-sum game
for rural families. Using rural-origin migrant children as the comparison group, children who
were left behind show significant and nontrivial disadvantages in terms of school performance.
They score approximately 28 percent of a standard deviation lower in cognitive abilities tests,
spend 2.5 percent of a standard deviation less effort on homework, and are less attached to
China’s Internal Migration and the Wellbeing of Rural-origin Children
26
schools by approximately 10 percent of a standard deviation. These findings are not surprising,
considering that the education resources and quality of schools in cities are much better than
those in rural areas. Similarly, we also observe a sizable gap in physical health between
left-behind children and migrant children by over 30 percent of a standard deviation. Better
health care facilities in cities and better care given by parents could both contribute to migrant
children’s superior health conditions. However, these two groups perform quite similarly in
terms of mental health and future aspirations, probably because they have all come across
psychological and developmental difficulties one way or another. To sum up, as the evidence
demonstrates, more migrant parents would bring their children to cities than not.
4. Does Local Hukou Payoff?
In addition to the financial and psychological costs associated with migration, migrant children
in urban China suffer additional disadvantages from the institutional barriers set by the hukou
system. An easy way to directly see the payoffs of obtaining a local hukou status is to contrast
those rural-origin children who successfully converted their hukou to those who failed to do so.
We would expect better school performance and future aspirations for the former group,
because they face far less institutional obstacles in cities. However, the propensity score
matching results in Table 6 show that except for some advantage in cognitive abilities for hukou
converters (arguably due to their greater chances of getting into high quality local schools), these
two groups actually do not differ significantly in most wellbeing measures (although the signs of
the effects are in line with our prediction). The relatively small sample size is of course a
limitation, but the other plausible explanation is that these converters are a fairly heterogeneous
group as it consists of children with very successful parents and also children originating from
ordinary rural families who were lucky enough to obtain local urban hukou through village
incorporation. Whatever the explanation is for the few differences between hukou converters and
non-converters, it suggests that the formal acquisition of a local urban hukou can only partially
compensate for the problems associated with migrating from rural to urban areas in China
(Zhang and Wu 2013).
China’s Internal Migration and the Wellbeing of Rural-origin Children
27
Conclusions and Discussion
With one-sixth of its population on the move and two-fifths of its children being directly affected,
China’s internal migration has been continually shaping the country’s future in a quiet but
irreversible manner. The wellbeing and life prospects of these rural-origin children have become
important public agendas for scholars and policymakers. Despite mounting interest in this topic,
relevant studies have been inconclusive, largely because they only focused on certain phases or
aspects of the whole migration process.
In this article, based on a nationally representative survey on Chinese junior high students in
both rural and urban areas, we developed a comprehensive analytic framework to investigate the
impacts of migration. Specifically, we identified three important steps of the migration process:
parental migration, child migration, and hukou conversion. Based on these steps, we classified
four groups of rural-origin children: hukou converters and migrants who live in cities, and
left-behinds and non-migrants who live in the countryside. Four comparison strategies were
employed accordingly for causal inferences, and hypotheses were examined on diverse child
wellbeing domains including school performance, physical and mental health, and future
aspirations. We assessed the impacts of child migration by contrasting migrants with
non-migrants, and the impacts of parental migration by contrasting left-behinds with
non-migrants. Moreover, we further contrasted left-behinds with migrants to evaluate the relative
benefits (or costs) of these two migration strategies, and contrasted hukou converters with
migrants to evaluate the value of obtaining a local hukou. As migration is a process full of
selection, to ensure the comparability of different groups of rural-origin children, we employed a
propensity score matching method. Specifically, optimal full matching was chosen for its
methodological superiority over other matching methods.
Results after propensity score matching demonstrated that both benefits and costs are
involved in the migration process. First, rural migrant children do benefit from better educational
resources and health care facilities offered in cities. Migrant children have significantly better
school performance and physical health than their non-migrant peers remaining in the
countryside. However, the migration experience per se, and possibly the negative peer influences
China’s Internal Migration and the Wellbeing of Rural-origin Children
28
resulting from hukou-based segregation, have greatly reduced migrant children’s educational
aspirations and increased their anxiety toward the future. On the other hand, although leaving
children behind in the countryside means they will not be exposed to the adverse influences from
the migration experience and institutional obstacles in cities, it does generate severely negative
impacts on their physical and mental health when they are separated from their parents for long
periods of time. Taken together, migrating with parents to cities does seem to be the preferable
choice, as migrant children clearly enjoy large premiums in school performance and physical
health.
Nevertheless, as pointed out before, a major limitation of our data is that only officially
enrolled children were sampled in this school-based survey. As informed by other research using
census data (Liang and Chen 2007; Wu and Zhang 2015), the dropout rate is significantly higher
for migrant children than left-behind children, and this will no doubt add to migrant parents’
concerns when deciding whether or not to bring their children to cities. Finally, even with the
relatively small sample size and large within-group heterogeneity, we were still able to observe
significant gains in academic achievement among hukou converters compared to the other rural
migrant children. An important mechanism, we presume, is less school segregation. By
successfully converting their hukou status, this group of rural-origin migrant children does not
face institutional discrimination when entering high quality local schools, resulting in better
academic achievements. In fact, we would even expect significantly a higher enrollment rate for
hukou converters than other rural migrants, although we not have data on school enrollment to
verify this hypothesis.
We therefore conclude that although leaving their children behind appears to be a less wise
choice for migrant parents, the many uncertainties and risks pertaining to the migration process,
as well as various institutional barriers posed by the hukou system, combine to strongly dissuade
parents from bringing their children along to destination cities. Obtaining a local urban hukou
may promote rural-origin children’s assimilation into local societies by offering them better
educational opportunities, but it does not shield them from the other negative side-effects of the
migration experience. The migration process poses both opportunities and challenges for rural
China’s Internal Migration and the Wellbeing of Rural-origin Children
29
families, and trade-offs are inevitable in various migration decisions. There are no simple
answers to complicated questions. Thus in the assessment of migration’s positive and negative
effects on rural children’s wellbeing, any over-generalized conclusion should be avoided.
China’s Internal Migration and the Wellbeing of Rural-origin Children
30
References
ACWF. 2013. "Woguo Nongcun Liushou Ertong, Chengxiang Liudong Ertong Zhuangkuang Yanjiu
Baogao. (Research Report on the Situation of Rural Left-Behind China, Rural-urban Migration
Children in China). http://acwf.people.com.cn/n/2013/0510/c99013-21437965.html#." vol. 2014.
Borjas, George J. 1987. "Self-Selection and the Earnings of Immigrants." The American Economic
Review 77:531-553.
Caliendo, Marco and Sabine Kopeinig. 2008. "Some Practical Guidance for the Implementation of
Propensity Score Matching." Journal of Economic Surveys 22:31-72.
Chan, Kam Wing and Will Buckingham. 2008. "Is China Abolishing the Hukou System?" The China
Quarterly 195:582-606.
Chan, Kam Wing and Li Zhang. 1999. "The Hukou System and Rural-Urban Migration in China:
Processes and Changes." The China Quarterly 160:818-855.
Chen, Xinxin, Qiuqiong Huang, Scott Rozelle, Yaojiang Shi, and Linxiu Zhang. 2009. "Effect of
Migration on Children's Educational Performance in Rural China." Comparative Economic Studies
51:323.
Dronkers, Jaap and Manon De Heus. 2013. "Immigrant Children’s Academic Performance: The Influence
of Origin, Destination, and Community." Pp. 247-265 in PISA, Power and Policy: The Emergence of
Global Educational Governance, edited by H.-D. Meyer and A. Benevot. Oxford: Symposium
Books.
Feliciano, Cynthia. 2005. "Educational Selectivity in U.S. Immigration: How Do Immigrants Compare to
Those Left Behind?" Demography 42:131-152.
Gao, Yang, Li Ping Li, Jean Hee Kim, Nathan Congdon, Joseph Lau, and Sian Griffiths. 2010. "The
impact of parental migration on health status and health behaviours among left behind adolescent
school children in China.(Research article)." BMC Public Health 10:56.
Goodburn, Charlotte. 2009. "Learning from Migrant Education: A Case Study of the Schooling of Rural
Migrant Children in Beijing." International Journal of Educational Development 29:495-504.
Guo, Shenyang and Mark W Fraser. 2014. Propensity Score Analysis: Statistical Methods and
Applications, vol. 11: Sage Publications.
Haviland, Amelia, Daniel S Nagin, and Paul R Rosenbaum. 2007. "Combining Propensity Score
Matching and Group-based Trajectory Analysis in An Observational Study." Psychological Methods
12:247.
Heckman, James J, Hidehiko Ichimura, and Petra E Todd. 1997. "Matching as An Econometric
Evaluation Estimator: Evidence from Evaluating A Job Training Programme." The Review of
Economic Studies 64:605-654.
Holland, Paulw. 1986. "Statistics and Causal Inference." Journal of the American Statistical Association
81:945-960.
Hu, Feng. 2012. "Migration, Remittances, and Children's High School Attendance: The Case of Rural
China." International Journal of Educational Development 32:401-411.
China’s Internal Migration and the Wellbeing of Rural-origin Children
31
Jia, Zhaobao and Wenhua Tian. 2010. "Loneliness of Left‐behind Children: A Cross‐sectional Survey
in A Sample of Rural China." Child: Care, Health and Development 36:812-817.
Lehmann, Erich Leo and Howard JM D'Abrera. 2006. Nonparametrics: Statistical Methods based on
Ranks. New York: Springer
Li, Qiang, Gordon Liu, and Wenbin Zang. 2015. "The Health of Left-behind Children in Rural China."
China Economic Review Forthcoming.
Liang, Zai and Yiu Por Chen. 2007. "The Educational Consequences of Migration for Children in China."
Social Science Research 36:28-47.
Liang, Zai, Lin Guo, and Chengrong Duan. 2008. "Migration and the Well-Being of Children in China."
The Yale-China Health Journal 5:25-46.
Lu, Yao. 2012. "Education of Children Left behind in Rural China." Journal of Marriage and Family
74:328-341.
Lu, Yao and Hao Zhou. 2013. "Academic Achievement and Loneliness of Migrant Children in China:
School Segregation and Segmented Assimilation." Comparative Education Review 57:85.
Mu, Ren and Alan Brauw. 2015. "Migration and Young Child Nutrition: Evidence from Rural China."
Journal of Population Economics 28:631-657.
Rosenbaum, Paul R. 2002. Observational Studies. New York: Springer.
South, Scott J., Dana L. Haynie, and Sunita Bose. 2007. "Student Mobility and School Dropout." Social
Science Research 36:68-94.
Wen, Ming and Danhua Lin. 2012. "Child Development in Rural China: Children Left behind by Their
Migrant Parents and Children of Nonmigrant Families." Child Development 83:120-136.
Wen, Ming, Shaobing Su, Xiaoming Li, and Danhua Lin. 2015. "Positive Youth Development in Rural
China: The Role of Parental Migration." Social Science & Medicine 132:261-269.
Wu, Qiaobing, Deping Lu, and Mi Kang. 2015. "Social Capital and the Mental Health of Children in
Rural China with Different Experiences of Parental Migration." Social Science & Medicine
132:270-277.
Wu, Xiaogang and Donald J. Treiman. 2004. "The Household Registration System and Social
Stratification in China: 1955-1996." Demography 41:363-384.
Wu, Xiaogang and Zhuoni Zhang. 2015. "Population Migration and Children’s School Enrollments in
China, 1990–2005." Social Science Research 53:177-190.
Xiong, Yihan. 2015. "The Broken Ladder: Why Education Provides No Upward Mobility for Migrant
Children in China." The China Quarterly 221:161-184.
Xu, Duoduo and Xianggang Wu. 2015a. "The Rise of the Second Generation: Aspirations, Motivations
and Academic Success of Chinese Immigrants' Children in Hong Kong." in PSC Research Report No.
15-852.
Xu, Duoduo and Xiaogang Wu. 2015b. "Separate and Unequal: Hukou-based School Segregation and
Educational Inequality in Urban China." Working Paper.
China’s Internal Migration and the Wellbeing of Rural-origin Children
32
Xu, Hongwei and Yu Xie. 2015. "The Causal Effects of Rural-to-Urban Migration on Children’s
Well-being in China." European Sociological Review 31:502-519.
Zhang, Zhuoni and Donald J. Treiman. 2013. "Social origins, hukou conversion, and the wellbeing of
urban residents in contemporary China." Social science research 42:71-89.
Zhao, Qiran, Xiaohua Yu, Xiaobing Wang, and Thomas Glauben. 2014. "The Impact of Parental
Migration on Children's School Performance in Rural China." China Economic Review 31:43-54.
Zuccotti, Carolina V, Harry BG Ganzeboom, and Ayse Guveli. 2015. "Has Migration Been Beneficial for
Migrants and Their Children?" International Migration Review Forthcoming.
China’s Internal Migration and the Wellbeing of Rural-origin Children
33
Appendix
Table A1. Naïve OLS Regression Models Predicting Differences in Various Outcomes between Migrant Children and Non-migrant Children
Treatment
(migrants=1)
Controls
Constant
Observations
R-squared
(1)
Cognitive
abilities
(2)
Hours/day spent
on homework
(3)
Attachment to
school
(5)
Health
(4)
Depression
(6)
Educational
aspirations
(7)
Confidence
about future
0.227**
0.111
-0.018
0.211***
0.041
-0.164
-0.108***
(0.070)
Yes
-0.278**
(0.100)
5,318
0.095
(0.097)
Yes
2.595***
(0.165)
5,027
0.037
(0.016)
Yes
2.036***
(0.032)
5,293
0.030
(0.042)
Yes
3.884***
(0.079)
5,317
0.044
(0.036)
Yes
2.096***
(0.073)
5,205
0.042
(0.127)
Yes
5.582***
(0.181)
5,318
0.073
(0.031)
Yes
3.189***
(0.064)
5,286
0.058
Notes: Robust standard errors adjusted for the clustering effect on schools are in parentheses; controls include age, gender, school grade, minority, region
of origin, grade retention during primary school, highest level of parental education, number of siblings, family finance situation before primary school,
and number of books at home; *** p<0.001, ** p<0.01, * p<0.05.
34
China’s Internal Migration and the Wellbeing of Rural-origin Children
Table A2. Naïve OLS Regression Models Predicting Differences in Various Outcomes between Left-behind Children and Non-migrant Children
Treatment
(left-behinds=1)
Controls
Constant
Observations
R-squared
(1)
Cognitive
abilities
(2)
Hours/day spent
on homework
(4)
Attachment to
school
(5)
Health
(3)
Depression
(6)
Educational
aspirations
(7)
Confidence
about future
0.025
0.112
-0.027*
-0.058
0.066*
0.076
-0.063*
(0.031)
Yes
-0.067
(0.119)
4,795
0.070
(0.077)
Yes
2.391***
(0.420)
4,550
0.052
(0.011)
Yes
1.933***
(0.130)
4,774
0.041
(0.031)
Yes
3.904***
(0.286)
4,794
0.034
(0.027)
Yes
2.146***
(0.154)
4,682
0.040
(0.095)
Yes
5.302***
(0.629)
4,795
0.051
(0.027)
Yes
2.801***
(0.161)
4,765
0.056
Notes: Robust standard errors adjusted for the clustering effect on schools are in parentheses; controls include age, gender, school grade, minority, region
of origin, grade retention during primary school, highest level of parental education, number of siblings, family finance situation before primary school,
and number of books at home; *** p<0.001, ** p<0.01, * p<0.05.
35
China’s Internal Migration and the Wellbeing of Rural-origin Children
Table A3. Naïve OLS Regression Models Predicting Differences in Various Outcomes between Left-behind Children and Migrant Children
Treatment
(left-behinds=1)
Controls
Constant
Observations
R-squared
(1)
Cognitive
abilities
(2)
Hours/day spent
on homework
(4)
Attachment to
school
(5)
Health
(3)
Depression
(6)
Educational
aspirations
(7)
Confidence
about future
-0.180*
-0.046
-0.017
-0.265***
0.022
0.184
0.024
(0.076)
Yes
-0.124
(0.111)
3,015
0.097
(0.129)
Yes
2.814***
(0.177)
2,839
0.051
(0.020)
Yes
2.032***
(0.035)
2,995
0.030
(0.046)
Yes
4.010***
(0.089)
3,015
0.070
(0.046)
Yes
2.110***
(0.079)
2,955
0.038
(0.149)
Yes
5.043***
(0.253)
3,015
0.079
(0.039)
Yes
3.100***
(0.070)
2,987
0.059
Notes: Robust standard errors adjusted for the clustering effect on schools are in parentheses; controls include age, gender, school grade, minority, region
of origin, grade retention during primary school, highest level of parental education, number of siblings, family finance situation before primary school,
and number of books at home; *** p<0.001, ** p<0.01, * p<0.05.
36
China’s Internal Migration and the Wellbeing of Rural-origin Children
Table A4. Naïve OLS Regression Models Predicting Differences in Various Outcomes between Hukou Converters and Migrant Children
Treatment
(converters=1)
Controls
Constant
Observations
R-squared
(1)
Cognitive
abilities
(2)
Hours/day spent
on homework
(3)
Attachment to
school
(4)
Health
(5)
Depression
(6)
Educational
aspirations
(7)
Confidence
about future
0.147
0.251
-0.018
0.079
0.023
-0.190
0.052
(0.086)
Yes
-0.201
(0.137)
1,919
0.119
(0.143)
Yes
2.640***
(0.192)
1,801
0.063
(0.029)
Yes
2.075***
(0.049)
1,907
0.023
(0.077)
Yes
4.101***
(0.106)
1,919
0.032
(0.075)
Yes
2.094***
(0.106)
1,887
0.045
(0.171)
Yes
4.551***
(0.351)
1,919
0.116
(0.062)
Yes
3.056***
(0.091)
1,901
0.075
Notes: Robust standard errors adjusted for the clustering effect on schools are in parentheses; controls include age, gender, school grade, minority, region
of origin, grade retention during primary school, highest level of parental education, number of siblings, family finance situation before primary school,
and number of books at home; *** p<0.001, ** p<0.01, * p<0.05.
China’s Internal Migration and the Wellbeing of Rural-origin Children
37
0
0
.1
.1
.2
.2
Density
Density
.3
.3
.4
.4
.5
.5
Histograms of Estimated Propensity Scores
-10
-5
0
Non-migrant Children
5
-10
-5
0
Migrant Children
5
-10
Propensity Scores
-5
0
5
Boxplots of Estimated Propensity Scores
Non-migrant Children
Migrant Children
Figure A1. Histograms and Boxplots of Estimated Propensity Scores for Migrant Children and
Non-migrant Children
China’s Internal Migration and the Wellbeing of Rural-origin Children
38
0
0
.5
.5
Density
Density
1
1
1.5
1.5
Histograms of Estimated Propensity Scores
-1
0
1
2
Non-migrant Children
3
-1
0
1
2
Left-behind Children
3
-1
0
Propensity Scores
2
1
3
Boxplots of Estimated Propensity Scores
Non-migrant Children
Left-behind Children
Figure A2. Histograms and Boxplots of Estimated Propensity Scores for Left-behind Children
and Non-migrant Children
China’s Internal Migration and the Wellbeing of Rural-origin Children
39
0
0
.2
.2
Density
Density
.4
.4
.6
.6
Histograms of Estimated Propensity Scores
-5
0
5
Migrant Children
10
-5
0
5
Left-behind Children
10
-5
Propensity Scores
0
5
10
Boxplots of Estimated Propensity Scores
Migrant Children
Left-behind Children
Figure A3. Histograms and Boxplots of Estimated Propensity Scores for Left-behind Children
and Migrant Children
China’s Internal Migration and the Wellbeing of Rural-origin Children
40
0
0
.1
.1
.2
.2
Density
Density
.3
.3
.4
.4
.5
.5
Histograms of Estimated Propensity Scores
0
2
4
6
Migrant Children
8
10
0
2
6
4
Hukou Converters
8
10
0
2
Propensity Scores
4
6
8
10
Boxplots of Estimated Propensity Scores
Migrant Children
Hukou Converters
Figure A4. Histograms and Boxplots of Estimated Propensity Scores for Hukou Converters
and Migrant Children
PSC Research Reports
The Population Studies Center (PSC) at the University of Michigan is one of the
oldest population centers in the United States. Established in 1961 with a grant
from the Ford Foundation, the Center has a rich history as the main workplace for
an interdisciplinary community of scholars in the field of population studies.
Currently PSC is one of five centers within the University of Michigan's Institute for
Social Research. The Center receives core funding from both the Eunice Kennedy
Shriver National Institute of Child Health and Human Development {R24) and the
National Institute on Aging (P30).
PSC Research Reports are prepublication working papers that report on current
demographic research conducted by PSC-affiliated researchers. These papers are
written for timely dissemination and are often later submitted for publication in
scholarly journals.
The PSC Research Report Series was initiated in 1981.
Copyrights for all Reports are held by the authors. Readers may quote from this
work (except as limited by authors) if they properly acknowledge the authors and
the PSC Series and do not alter the original work.
Population Studies Center
University of Michigan
Institute for Social Research
PO Box 1248, Ann Arbor, Ml 48106-1248 USA
www.psc.isr.umich.edu