See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/321320366

The Effect of Family Size on Children's Education: Evidence from the Fertility
Control Policy in China

Article  in  Frontiers of Economics in China · January 2017

DOI: 10.3868/s060-006-017-0003-3

CITATIONS READS

3 3,327

1 author:

Ying Shen
University of Notre Dame
4 PUBLICATIONS   13 CITATIONS   

SEE PROFILE

Some of the authors of this publication are also working on these related projects:

The Effect of Family Size on Children's Education View project

All content following this page was uploaded by Ying Shen on 27 November 2017.

The user has requested enhancement of the downloaded file.

Front. Econ. China 2017, 12(1): 37–65
DOI 10.3868/s060-006-017-0003-3

RESEARCH ARTICLE
 

Ying Shen

The Effect of Family Size on Children’s

Education: Evidence from the Fertility Control
Policy in China
Abstract Empirical research on the effect of family size on child education is
complicated by the endogeneity of family size. This study exploits plausibly
exogenous changes in family size caused by China’s population control policy to
estimate the causal relationship between family size and child education
outcomes. The results show that, compared to an only child, a person with an
additional sibling will have an approximate seventeen percentage points lower
likelihood of completing middle school in China. Separate regressions across
individual characteristics reveal that much of this negative effect appears to be
driven by the cohorts born in earlier years after the policy, and children with the
highest birth order within a family.

Keywords education, fertility control policy, family size

JEL Classification I20, J13, J18

1 Introduction
The relationship between family size and child quality, in particular as measured
by education outcomes, has captured economists’ attention in recent years (Black
et al., 2005; Moav, 2005; Lee, 2006; Qian, 2013). One reason for the prominence
is that education is a cornerstone of national development, and developing
countries that have much higher fertility typically have a lower education level
than developed countries. Moreover, even within the set of developing countries,
some countries that have achieved successful economic booms in the past few

Received August 17, 2015

Ying Shen ( )
Institute for Economic and Social Research (IESR), Jinan University, Guangzhou 510630,
China; Department of Economics, University of Notre Dame, Notre Dame, IN 46556, USA
E-mail: [email protected]
38 Ying Shen

decades—notably, China—have dramatically restrained population growth

during the same period. From an individual’s standpoint, education plays a great
role in determining career outcomes and living standards. Therefore, identifying
the effect of family size on children’s education is particularly important for
policy makers in order to formulate appropriate policies toward achieving
balanced socio-economic development, and also for parents who are
contemplating the child quantity-quality tradeoff.
Despite the importance of this topic, several theoretical and empirical
challenges complicate the inference of causal effects. In theory, the relationship
between family size and children’s educational attainment is inconclusive.
Family size can affect children’s education outcomes through two main channels.
The first channel is a “dilution effect,” where an increase in the number of
children will dilute parental resources being allocated on each child (Becker and
Lewis, 1973; Black et al., 2005; Lee, 2006). In this sense, the effect of family
size on each child’s education outcome is negative. By contrast, the second
channel is a “peer effect” or “spillover effect.” Some child psychologists such as
Iacovou (2001) point out that children can affect each other by learning-by-doing,
in which case more siblings provide children with more chances to teach and
learn from each other. Such spillovers, therefore, can lead to a positive
relationship between family size and child education outcomes. Given that these
two channels work in opposite directions, the overall effect of family size
remains ambiguous. In addition to this theoretical indeterminacy, empirical
endogeneity poses another big challenge in identifying causality as child
education and quantity are jointly determined by parents. For example, in most
developed countries like the United States, wealthier families who value
education are more likely to have fewer children, and, therefore, an OLS estimate
will overestimate the “dilution effect.” Alternatively, in some other countries,
parents who value education prefer to have more children, and so an OLS
estimate will underestimate the “dilution effect.” Therefore, in order to overcome
the endogeneity bias, an instrumental variable strategy is useful.
The correlation between family size and children’s education is not a new
research topic, and a number of existing empirical studies have been done in
related fields by economists and other social scientists. For instance, Lu and
Treiman (2005) find a small negative effect of sibling size on educational
attainment measured by years of schooling in China, though the effect is very
heterogeneous across cohort, residential area, sex and parental characteristics,
The Effect of Family Size on Children’s Education 39

however, in their paper they do not use any instrument or natural experiment
strategy. In recent work, many economists try to tackle the issue of bias by
implementing instrumental variables to take advantage of exogenous variation in
family size. For example, Black, Devereux and Salvanes (2005) use the twins
instrument and Lee (2006) uses a preference for sons as an instrument.1 Both of
the resulting papers find negligible negative effects of family size on child
education. However, in practice, Black uses data from Norway, and Lee uses data
from South Korea: The results in these developed countries may not be accurate
generalizations for developing countries.
There are two closely related papers that investigate the quantity-quality
tradeoff within a family using the One Child Policy in China as an exogenous
source of the variation in family size. Qian (2013) uses the relaxation of the One
Child Policy in rural China as a natural experiment,2 and her results show that an
increase in family size will significantly increase school enrollment for the first
child in rural families. However, since the relaxation of the One Child Policy her
research applied only to rural residents whose first child was a girl, her paper
does not include urban Chinese families; hence, the cross-sectional variation in
her paper is limited to rural areas, and does not apply to the country as a whole.
Moreover, while her dependent variable of interest is primary education
enrollment, family size could have different effects on other education outcomes.
Li (2011) employs the provincial differences in the implementation of the One
Child Policy as an instrument variable (IV) and shows a negative effect of family
size on the first-child’s educational attainment. However, due to the data
limitations, he could not distinguish the quantity-quality tradeoffs between urban
and rural sectors. Additionally, similar to Qian (2013), the study is only limited to
the enrollment outcomes of children, but measures of other educational
attainment such as completing middle school are also useful measures of
success.
This paper contributes to the existing studies in several ways. First, in order to
isolate the causal effect of family size on children’s education, I exploit both
region and birth year variation in the implementation of the compulsory fertility

1
Lee argues that Asian countries prefer boys to girls, so he uses the gender of the first child as
an instrumental variable.
2
On April 13, 1984, the Central Party Committee issued “Document 7,” which includes,
significantly, a relaxation known as the “1-son-2-child” rule that allowed rural couples to have
a second child if the first child was a girl.
40 Ying Shen

control policy.3 Specifically, I use the enforcement of the compulsory fertility

control policy in China interacted with the region of birth (urban or rural) as an
instrument to generate any change in the number of children per family, since the
policy is not equally enforced across different regions. By using both cross
sectional and time series variation in family size, I control for other time varying
factors that would potentially influence the outcomes. What is more, among the
papers which use the family planning policy in China as a natural experiment, to
my knowledge, this is the first to take the age of mothers into consideration.
Second, besides the mean effect that is over all groups of children, I try to
estimate various heterogeneous effects across variations in individual
characteristics, and I explore the mechanisms underlying these effects. Finally,
this study provides informative results for recent population policy discussions
and offers guidance for future policy-decision making. While a large number of
studies have attempted to estimate the effects of the family planning policy on
fertility and other health outcomes, very few papers (for example, Yang, 2006)
focus on the effect of this large scale family planning policy on children’s
education outcomes. This paper will enrich the assessments and educational
implications of the family planning policy.
By applying an exogenous institutional shock in family size to a dataset that
contains household and individual’s characteristics, I uncover several significant
results. First, like much of the previous literature, I find a negative relationship
between family size and child education measured by a child’s likelihood of
completing middle school. The IV estimates are higher in magnitude than
standard OLS estimates which are subject to endogeneity problems, and the
results are robust to several sensitivity checks. Second, the empirical results
show that the negative effects are mainly driven by the cohorts born in earlier
years after the policy, and children with the highest birth order within a family,
which confirms the existence of a “dilution effect.”
The remainder of this paper is organized as follows: Section 2 provides a brief
description of the historical context of family planning policies in China. Section
3 explains the construction of the data set and presents descriptive statistics.
Section 4 outlines the empirical identification strategies. Section 5 reports the
results and robustness checks. Section 6 concludes.

3
In this paper, I use the compulsory fertility control policy and the family planning policy
interchangeably.
The Effect of Family Size on Children’s Education 41

2 Background on Fertility Control Policies in China

From the establishment of the People’s Republic of China in 1949 until the 1960s,
the guiding ideology of population policy in China was to promote fertility. The
Chinese government encouraged families to have as many children as possible,
which resulted in a remarkable increase in the total Chinese population during
that period.4
However, beginning in the early 1970s, the Chinese government began to shift
its population policy from pro-fertility to fertility reduction. In practice, in
December 1973, the policy promoting “later [childbearing age], longer [time
between births], fewer [number of children]” was initially launched. When, in
February 1975, in the report for the national economy, Chairman Mao stated “the
growth of population has to be controlled,” enforcement of compulsory fertility
control gradually tightened across the country. The One Child Policy was
formally announced in 1978 and was written into the national constitution.
During the implementation process, females were limited to only one birth, and,
households were subject to a number of penalties if they violated the policy.
Local governments at all levels were given the responsibility to fulfill birth
quotas. In April 1984, in view of the difficulties of enforcement and social unrest
in rural areas, the central government issued a document which allowed rural
families to have a second child if their first child was a girl.5
In this paper, I will use the shift from pro-fertility to fertility control as an
exogenous variation in family size for analysis. It is obviously difficult to
pinpoint a specific date when the family planning policy began to affect family
size since the fertility control was initially promoted in December 1973 but was
officially written into the national constitution in 1978. However, in the early
years after fertility control was initially proposed, it was only an incentive policy;
but fertility control was tightened to a compulsory policy following Chairman
Mao’s 1975 statement, this empirical analysis, therefore, assumes this date as the
starting point for enforced fertility control policy.6
Although very restrictive in general, the federal family planning policy shows

4
Chinese population reached 850 million at the beginning of the 1970s.
5
See footnote 2.
6
If I assume 1978 as the starting point for the family planning policy, it is very likely that the
results will be contaminated by the differences in pre-trends since the family planning policy
has already taken effect even before it was formally written into the national constitution.
42 Ying Shen

great heterogeneity across race and region. First, most ethnic minorities,7 who
comprise about 8% of the total population of China, are exempted from this
policy. Within ethnic groups with a total population less than 10 million, women
were permitted to have more than one child.8 In addition to this Han-Minority
difference in the family planning policy, another great difference existed between
urban and rural areas, and could be attributed to two main reasons. First, as
described above, the Chinese government eventually loosened the restrictions of
the family planning policy on rural families to allow them one more child if their
first was a girl. Because this relaxation was not applicable to urban couples, the
policy itself applies differently across urban and rural areas. Second, the severity
of penalties for above-quota births varied, with urban residents facing much
stiffer penalties than rural residents for violation of the family planning policy.
Urban couples who had an above-quota number of children had to pay a fine as
high as 70 percent of their monthly salary, and could lose promotion
opportunities—or even their jobs, if they worked in government or state-owned
companies—and any above-quota urban children were denied access to public
services, such as public schools. By contrast, the only penalty for rural couples
was a one-time fine. Although the fine still represented a significant amount of
money for rural people, their opportunity cost of having extra children are lower
than that of their urban counterparts.
The heterogeneity of the family planning policy at the urban/rural level
provides a unique natural experiment to identify the effect of the compulsory
fertility control on children’s education outcomes. Although the overall fertility
trends of urban and rural families are both declining after the policy,9 urban
people are more affected, so if the urban children’s education outcomes are
suddenly altered after the birth control policy relative to rural children’s, the
variations in the education outcomes can be at least partially attributed to the
effect of the policy.

7
In addition to Han, the ethnic majority, the People’s Republic of China officially identifies
55 ethnic minorities which comprise 8.49% of the population of Chinese mainland.
8
In the late 1980s, only the Zhuang minority had a population of more than 10 million, so
Guangxi province, where most Zhuang people live was included in the implementation of the
family planning policy; however, because none of the other ethnic minorities reached this
number, families were allowed to have several children, regardless of where they lived.
9
See Figure 1.
The Effect of Family Size on Children’s Education 43

3 Data and Summary Statistics

To implement my empirical strategy, I use microdata from the 2008 Rural-Urban
Migration in China and Indonesia Survey (RUMiCI). 10 The 2008 RUMiCI
survey in China is conducted in ten provinces, including Henan, Jiangsu, Sichuan,
Hubei, Anhui, Shanghai, Zhejiang, Guangdong, Hebei and Chongqing. The entire
sample size covers approximately 18,000 households in China, consisting of
5,000 migrant, 5,000 urban and 8,000 rural households. The survey contains
information on each individual’s year of birth, region of residence, sex, ethnicity,
educational attainment, and relationship to the head of household, etc. Because I
can match children’s information to their parents’ by household ID code, the data
provides an opportunity to calculate family size and rank each child’s birth order.
I restrict the sample by the following steps. First, since ethnic minorities could
have exemptions from the family planning policy, I restrict my attention to Han
people. Because this ethnic majority exceeds 92% of China’s total population,
the exclusion of ethnic minorities does not significantly influence my results. I
further restrict my sample to individuals who were born from 1950 to 1990, or
equivalently, individuals aged 18 to 58 in 2008. In addition, I exclude migrant
families (those whose household are neither urban nor rural in the surveys),
because it is difficult to judge the level of policy enforcement faced by migrants.
However, in my robustness check, I include migrants by treating them as urban
residents if their Hukous11 are registered to urban places and rural residents if
their Hukous are registered to rural places, and the results show robustness to this
particular sample selection criterion. Finally, my cut-off age for mothers is 30

10
The Longitudinal Survey on Rural Urban Migration in China (RUMiC) consists of three
parts: the Urban Household Survey, the Rural Household Survey and the Migrant Household
Survey. It was initiated by a group of researchers at the Australian National University, the
University of Queensland and Beijing Normal University and was supported by the Institute
for the Study of Labor (IZA), which provides the Scientific Use Files. The financial support
for RUMiC was obtained from the Australian Research Council, the Australian Agency for
International Development (AusAID), the Ford Foundation, IZA and the Chinese Foundation
of Social Sciences.
11
A Hukou, also called Huji, is a record in the system of household registration that originated
in ancient China, and has been required by law in the People’s Republic of China since the
1950s. Its aim is to regulate and tightly control the movement of people between urban and
rural areas. In the 1990s, the government relaxed the Hukou system and allowed rural residents
to apply for “temporary urban residency permits” to legally work in urban cities; however,
people who live outside their Hukou domains do not qualify for local benefits, including health
care, social security, education, employment, etc.
44 Ying Shen

years old12 in the year 1975 for individuals who were born before 1975. The age
of mothers matters because the family planning policy may generate different
effects for mothers at different ages. To make this idea concrete, consider two
mothers, one who was 40 years old in 1975, and the other one who was only 20.
For the mother who was 40, with or without the policy, she was unlikely to
continue bearing children, due to natural aging. Therefore, the policy will have
had little impact on her. However, for the mother who was only 20, and in her
prime child bearing years, the policy would have had a much bigger effect.
Based on the restrictions outlined above, I divide the sample into four mutually
exclusive categories: rural cohorts born before 1975 with older mothers;13 urban
cohorts born before 1975 with older mothers; rural cohorts born after 1975, or
born before 1975 with younger mothers;14 and urban cohorts born after 1975, or
born before 1975 with younger mothers. The first two categories are the
unaffected group and the last two categories are the affected group. Table 1

Table 1 Summary Statistics (Mean and Standard Deviation)

Urban area Rural area
Variable
Unaffected Affected Unaffected Affected
Age 40.5719 26.5326 39.4103 24.7052
(5.3674) (4.0190) (4.8909) (4.9465)
Married 0.8058 0.2209 0.9103 0.3437
(0.3963) (0.4151) (0.2863) (0.4750)
Number of Siblings 1.4244 0.3398 2.1065 1.4952
(1.1271) (0.6607) (1.4498) (1.0418)
Household head years of 9.5556 9.9015 5.7962 7.3189
schooling (4.0780) (2.9519) (2.1572) (2.2738)
Mother’s age 68.7359 53.5241 67.5932 50.1994
(4.0780) (5.3213) (4.9121) (6.1501)
Mother’s age at her first 28.6698 26.6217 28.4068 25.5105
delivery (4.0112) (3.6191) (5.2172) (4.4118)
Middle school 0.7194 0.9280 0.7308 0.8727
completion rate (0.4501) (0.2586) (0.4443) (0.3334)
N 278 1,181 312 5,552
Note: Affected groups are those either born after 1975, or born before 1975 with younger
mothers. People who finish middle school are defined as those whose years of education are
greater than nine or whose highest education level is bigger than or equal to seven. N
represents the number of individuals.

12
I set the cut-off age at 30 years old because, first, the fertility rate begins to decline after 30
years old and, second, in the sample from 1950 to 1975, over 85% of women deliver their last
babies before 30 years of age. Furthermore, in the opinion of Chinese people, 30 years old is
considered as the beginning of an advanced age for maternity.
13
Older mothers are those who were no younger than 30 years old in 1975.
14
Younger mothers are those who were younger than 30 years old in 1975.
The Effect of Family Size on Children’s Education 45

displays the summary statistics for the sample, with decomposition by region of
household residence and year of birth for each individual. The big difference in
marriage rates before and after 1975 reflects the much younger relative ages of
those married after that year.
Family size measure is considered as completed family size. Because I exclude
individuals who were younger than eighteen in 2008, as previously explained, I
can be fairly certain when family size is “complete” since there is little
possibility that a child of 18 has another sibling who is not accounted for in the
survey. I base determination of birth order on the straightforward chronology of
each child’s year of birth. In terms of educational attainment, I am interested in
whether or not a person completes a middle school education, a milestone
identified by an education level code no less than seven, or greater than nine
years of formal education. The average middle school completion rates of both
urban and rural people are provided in Table 1 and Figure 2. From the table, we
can see that middle school completion rates increased by 20% and 14% in urban
and rural areas, respectively.

Figure 2 Middle School Completion Rates by Urban and Rural Areas

Note: The middle school completion rate is smoothed with a three year moving average.

One potential problem that can arise with Chinese data on family size is that
parents may intentionally misreport the total number of children they have in
46 Ying Shen

order to avoid the penalties of violating the family planning policy. This concern
is not unreasonable due to the severe punishments for violation. However,
previous literature on the comparison of hospital birth records and population
census data shows that the misreporting problem is only significant for children
younger than two years of age (Zeng et al.,1993). Moreover, the population
census, around every ten years, encourages accurate reporting through incentives,
such as a lesser than usual fine for above-quota children to get the Hukou. My
use of data from the 2008 survey examines only children older than eighteen
years old, and so ensures the misreporting problem will not be atypically large in
my study.

4 Estimation Strategy
4.1 Empirical Framework

I use the following linear probability model to estimate the general effect of
number of siblings on the education outcome of interest of each individual:
Yitrp = β0 + β1Siblingsitrp + X itrp β 2 + β3 Z tp + Trendtp + γ t + ϕ p + δ r + ε itrp . (1)

In this specification, the education outcome is the probability of completing

middle school education. The dependent variable is a dummy variable which
equals one if the individual finishes middle school and zero otherwise. The
education outcome for individual i, born in province p, birth year t, birth order r,
is a function of: Siblingsitrp, the number of siblings; Xitrp, a set of demographic
characteristics including gender, household head’s education, rural or urban
designation of residence, and mother’s age relative to thirty years in 1975; Ztp,
the total number of full time middle school teachers in each province,15 which is
used for controlling changes from the supply side; to account for the possibility
that some aggregate unobservables might affect the education outcome, I also
include province-specific linear cohort time trends Trendtp. γ t , year of birth fixed
effects, which controls for non family size birth year effects in a better way than

15
It denotes the total number of middle school teachers in each province as of the time the
individual was age 12 since 12 is the normal age for a child to begin middle school. For
example, if an individual was born in 1980, I use the number of middle school teachers in
1992 in the province of living to control for the education supply shock. The data is coming
from China Compendium of Statistics 1949–2004.
The Effect of Family Size on Children’s Education 47

just having a dummy for being born after the policy; ϕ p , ten provinces fixed
effects, which controls for cross-province differences; and δ r , birth order fixed
effect, allowing for different effects at different birth orders. For all regressions,
standard errors are robustly clustered at the provincial level, although results of
using other clustering methods will also be reported in the robustness checks.
The simple OLS regression may have several problems: first, children born in
more recent years are more likely to have a higher probability of completing
middle school; second, the number of children is not exogenous, but rather
determined by parental preferences and correlated with many unobservable
variables. For instance, suppose the relationship between number of siblings and
children’s education is negative, and wealthier families who value education
choose to have fewer children. In that case, the OLS estimates will overestimate
the negative effect of an extra sibling on educational attainment. By contrast, if
the poorer families who value education less choose to have fewer kids, then the
OLS estimates will underestimate the negative effect. As a result, the OLS
estimates will be ambiguously biased. Controlling for birth year fixed effects
addresses the first concern to some extent; however, in order to overcome the
endogeneity issue, I exploit an instrumental variable strategy. The instrument for
the number of siblings is the interaction of two binary variables, one indicating
whether or not an individual is in the affected group, and the other indicating
whether or not the household is living in urban areas.
Therefore, the first-stage equation that relates the endogenous regressor to the
instrument and other control variables is:
Sibilingsitrp = α 0 + α1 Affected t * Urbani + X itrp α 2 + α3 Z tp + Trendtp
(2)
+ γt + ϕ p + δ r + μ itrp .

In this equation, the number of siblings for individual i, born in province p,

birth year t, birth order r, is a function of: the interaction term of Affected and
Urban. Affected, is one if an individual was born after 1975, or before 1975 with
a younger mother; and Urban, is one if the household is living in urban areas. All
other demographic characteristics and fixed effects are the same as in equation
(1). Since I have controls for an urban location and a mother’s age in 1975, and a
set of year dummies, my specification will only rely on the interaction of Urban
and Affected.
48 Ying Shen

4.2 Validity of the Instrumental Variable

Table 1 and Figure 1 demonstrate that family size has been significantly
decreased after the compulsory fertility control policy, especially for urban
families; however, in order for the 2SLS estimates to be unbiased and consistent,
several concerns should be ruled out.

Figure 1 Average Number of Siblings of Each Person by Urban and Rural Areas
Note: The number of siblings is smoothed with a three year moving average.

4.2.1 Correlated with Unobservables

The first necessary condition for a valid instrumental variable is that the variable
should be uncorrelated with other unobservables. Although this is fundamentally
improvable, given the fact that the compulsory fertility control policy is a
widespread and strict policy enforced among people across all occupations, all
income and education levels, and all social backgrounds (except for the
previously noted disparities between urban and rural, and between Han and other
minorities), the relationship between the implementation of the policy and other
individual unobservable characteristics should be insignificant.16 Of course, in
addition to any potential individual unobservables, the IV could also be

16
After the dawn of the 21st century, as enforcement of the One Child Policy has been
relaxed, high income people who can afford the fine and can send their above-quota children
to international schools are likely to have more than one child; however, at least before 1990,
the One Child Policy is generally equally enforced across education and income groups.
The Effect of Family Size on Children’s Education 49

correlated with some aggregate unobservables. However, given that in the

baseline model, I have already included province-specific cohort time trends,
such concern could be attenuated.

4.2.2 Intention to Be Treated

The second concern about the IV is whether or not the family planning policy is a
completely exogenous event. Monetary rewards to one-child families might be
seen to give families an incentive to have only one child, and, thereby, bias the
estimate. In reality, the monetary subsidy to one-child families is a very small
amount relative to earnings. To take Henan province as an example, in 2004,
although the annual subsidy for one-child families increased from 30 yuan17 to
120 yuan, the annual per capita disposable income remained significantly higher,
at 9,421 yuan for urban citizens, and 2,936 yuan for rural citizens.18 What is
more, Chinese people traditionally believe “the more children a family has, the
merrier the life” (duo zi duo fu), and perceive the benefits of having more
children as much greater than the monetary subsidy. Therefore, monetary
incentive will not significantly impact family size outcomes.

4.2.3 Education Policies

Another concern is that the IV may not satisfy the exclusion restriction if other
policies exist, in addition to the fertility policy, which also occurred differentially
between urban and rural areas, and have impacted educational attainments. This
condition, too, is inherently untestable. Given that my dependent variable of
interest is the probability of finishing middle school, and the year of birth spans
from 1950 to 1990, I will mainly look at the Compulsory Education Law. In 1986,
the Compulsory Education Law was formally promulgated, requiring that all
citizens 6–16 years old complete at least nine years of education. The three main
features of this compulsory education are continuity, proximity and
uniformity 19 —aim to achieve equally universal basic education across all

17
The exchange rate between US dollar and Chinese yuan is around 6.25 in 2014.
18
Data source: China Compendium of Statistics 1949–2004.
19
Continuity means students finish both primary and middle school education; proximity
means students can go into their nearest school without taking a middle school entrance exam;
uniformity means schools practice unified management in school administration, teaching and
education.
50 Ying Shen

locations and ethnicities in China. 20 Therefore, even though compulsory

education is not totally free in China, its compulsory nature, goal of equality, and
three features legally ensure that all children in China have equal opportunity to
receive nine years of basic schooling.
Despite the rapid development in both urban and rural basic education,
followed by the introduction of compulsory education, a disparity in education
across regions might still exist, 21 since the responsibility of popularizing
compulsory education is assigned to local governments, and each region has
different levels of economic development. Disparities could also be attributed to
a gap in teacher quality and quantity, or in investments between urban and rural
areas. 22 To see whether these gaps exist, it would be ideal to examine
information about teacher quality and compulsory education expenditures in
rural and urban areas from 1986 to 2006.23 Although I cannot find a source for
this complete range of data, I have included in the appendix such information
from 1996 to 2003. Figures A1 (a) and A1 (b) provide the student-teacher ratio in
urban schools, and the national average in primary and junior secondary schools,
respectively; Figures A2 (a) and A2 (b) report the education expenditures per
student in rural schools, and the national average by type of education; and
Figure A2 (c) reports the total education expenditures in compulsory education in
urban and rural areas. These figures show that, at least after 1996, rural
compulsory education faces a worse environment than urban compulsory
education in levels, but not in differential trends. Furthermore, Li (2009) shows
that due to the lack of teachers and investments in rural schools, inequalities
between urban and rural areas are much greater for higher education than for
compulsory education levels.

4.2.4 Channels Other than Family Size

The final concern for the validity of the IV is that the compulsory fertility control

20
Date source: China’s Progress Towards the Millennium Development Goals, 2013 Report.
21
Inequalities exist across provinces and across urban and rural areas within a province.
22
According to the Ministry of Education, the proportions of middle school teachers with
university degree and above in urban and rural areas are 23.51 and 9.35, respectively, and the
student-teacher ratio is much higher in villages than in cities.
23
The last group affected by the Compulsory Education Law in my sample is those who were
born in 1990 and were 16 years old in 2006.
The Effect of Family Size on Children’s Education 51

policy could affect education outcomes through channels other than family size,
such as changes in income, labor market conditions, or people’s attitudes towards
education; it is, therefore, difficult to see how specification can separate all these
effects from that of family size. In this paper, I use cross-sectional variation to
control for non-sibling size-related effects of the policy, but this only partially
addresses the problem. Estimates could also be biased if something differentially
affected people in urban versus rural areas for the cohort born after 1975.
One possible scenario is that education became more valuable to people in
urban areas after 1975, and so urban parents became more likely to invest in
children’s education. Nevertheless, in the long history of China, both urban and
rural Chinese parents have placed a very high value on education, as education
has long been recognized as the critical way to get ahead in China. 24 Therefore,
most parents are willing to make great sacrifices to let their children acquire
higher and better education, regardless of where they live. The results of a survey
conducted in 2008 about Chinese rural parents’ attitudes towards their children’s
education show that more than 73% of Chinese rural parents place considerable
value on the next generation’s education.
One more possible obstacle to my empirical strategy is that policy might affect
education by addressing problems apart from family size. In particular, the
results may be driven by the faster income growth rate in urban areas or more
wealth owned by urban families. For example, if, following the compulsory
fertility control policy, urban parents tend to delay their childbearing ages until
they have better economic conditions, then the 2SLS will overestimate the
dilution effect of family size on educational attainment. Table 1 reports the
average mother’s age at her delivery, and the results reflect that mothers’ delivery
ages are quite constant for both urban and rural women both before and after the
policy, so this scenario is unlikely. A second way to take a family’s economic
condition into consideration is to control for income growth rates. Although I am
unable to obtain information concerning family-level income over time, in the
robustness check section I am able to control for regional level income growth
rates, 25 and the estimates appear to be consistent with my basic results. Also, in
the robustness checks, I add urban/rural cohort time trends to account for the

24
Especially in rural China, education is considered as the only possible way to escape
poverty and have a dignified life.
25
Regional level refers to province level by location (urban or rural).
52 Ying Shen

possibility that some unobservables differentially affect urban and rural areas that
also affect the child education outcome.
Taken together, the compulsory fertility control policy significantly decreased
the family size of urban people relative to rural people. Given that the policy is a
federal compulsory policy, it is unlikely to be correlated with either individual
unobservables or aggregate unobservables. Chinese people have no intention to
be treated during the implementation of the policy. Other non-sibling size-related
effects of the policy that might affect education outcomes have been separated
out. Research suggests that no other education related issues have occurred
differentially between urban and rural areas, and all of this suggests that the
identification strategy is sound.

5 Empirical Results and Robustness Checks

5.1 The Effect of the Compulsory Fertility Control Policy on Family Size

In Figure 1, we can intuitively see declining trends in family size for both urban
and rural families, following the institution of the fertility control policy, and that
the effect on urban people is significantly larger. Specifically, after the policy,
urban people tend to have zero siblings, which means most of them are the only
child in their families; by contrast, rural people are still likely to have about one
sibling, so the difference in family size between urban and rural families is one.
However, the relationship in the figures might be affected by many other
influences. In this section, I attempt to systematically test whether the family
planning policy has had an effect on family size, and measure the magnitude of
this effect. The estimates of the first stage equation (2) are reported in Table 2.
The dependent variable is the number of siblings for each person. Column (5) in
Table 2 is my preferred specification, where I include a set of individual, family
and province control variables, as well as a series of fixed effects and
province-specific linear cohort time trends. There is a negative and significant
result for the instrument, an Affected dummy variable interacted with an Urban
dummy variable, which suggests that the restriction on family size led to a
relative decline in the number of siblings an individual has by 0.36 for urban
residents after the policy was passed. The F-statistics in the last column is 10,
suggesting that the weak instrument is not a particular problem; however, in
order to alleviate the worry that the estimated standard errors might be sensitive
The Effect of Family Size on Children’s Education 53

to the clustering methods, in the robustness checks, I will report the results by
clustering standard errors at different levels.

Table 2 The Effect of Family Planning Policy on the Number of Siblings per Person
Independent Dependent variable: number of siblings
variable (1) (2) (3) (4) (5)
Affected*Urban –0.8929*** –0.8246*** –0.4295*** –0.3638*** –0.3641***
(0.1295) (0.1220) (0.1194) (0.1158) (0.1168)
Demographic
Yes Yes Yes Yes Yes
controls
Supply side
Yes Yes Yes Yes Yes
control
Province fixed
No Yes Yes Yes Yes
effect
Date of birth
year fixed No No Yes Yes Yes
effect
Birth order fixed
No No No Yes Yes
effect
Province-specific
No No No No Yes
cohort trends
F-statistics 48 46 13 10 10
N 6,921 6,921 6,921 6,921 6,921
Note: Standard errors are clustered at the provincial level and reported in parentheses.
*** P<0.01, ** P<0.05, *P<0.10. Demographic characteristics include gender, household
head’s education level, mother’s age in 1975 indicators and residence type (urban or rural).
The supply side control is the total number of full time middle school teachers in each
province. Province-specific linear cohort trends are also included. Fixed effects include the
dummies for province, date of birth year, and birth order fixed effects.

5.2 The Effect of Family Size on Children’s Education

5.2.1 Average Effect

To estimate the effect of family size on educational attainment, Table 3 presents

the second stage results. Column (1) represents the results from the OLS
regression, and column (2) represents the results from the 2SLS regression. The
results suggest that the IV coefficient differ significantly from the OLS
coefficient. The effect on children’s education is much larger after instrumenting
family size, and the effect could be interpreted as follows: all else being equal, an
additional sibling leads to a seventeen percentage points decline in the likelihood
54 Ying Shen

that an individual finishes middle school.

Table 3 Second Stage: Average Effect of Family Size on Children’s Education

Dependent variable: Probability of completing middle school
Independent variable OLS 2SLS
(1) (2)
–0.0231* –0.1786***
Number of siblings
(0.0113) (0.0579)
Demographic controls Yes Yes
Supply side control Yes Yes
Province fixed effect Yes Yes
Date of birth year fixed
Yes Yes
effect
Birth order fixed effect Yes Yes
Province-specific cohort
Yes Yes
trends
N 6,921 6,921
Note: Standard errors are clustered at the province level and reported in parentheses.
*** P<0.01, ** P<0.05, *P<0.10. The “number of siblings” is instrumented by Affected*Urban
in column (2). All control variables and fixed effects are the same as in Table 2.

There are two explanations for why the OLS result is much smaller than the
2SLS result in absolute value. First, the OLS estimate uses all the variations in
family size whereas the 2SLS estimate uses only the variation in sibling size
from zero to one. If the local average treatment effect (LATE) on education
outcome for individuals affected by the instrument differs significantly from that
of all other people, then the coefficient in the OLS regression will differ from
that in the 2SLS regression. Therefore, it could be the case that the largest effect
of sibling size on children’s education is the difference from zero to one, so the
2SLS yields a higher estimate than the OLS. The second explanation would be
due to the possibility that, in rural China, families who can afford better
education tend to have more children. Although this is not the normal
phenomenon in many developed countries, it is not unreasonable in China. As I
mentioned in section 2, even though rural families face a lower opportunity cost
for having above-quota children than do urban families, they will still be fined a
significant amount of money. Therefore, if poorer families can afford neither the
fine nor the associated costs of raising children, they are less likely to have many
children.
The Effect of Family Size on Children’s Education 55

5.2.2 Heterogeneous Effects across Cohort

Because the average effect may hide substantial variation by individual and
family characteristics, here I first estimate separate models by cohort. While the
unaffected group is the same for each regression, I divide the affected group into
two cohorts: The first cohort is individuals born before 1985; and the second
cohort is individuals born between 1985 and 1990. The IV regression results of
each cohort are shown correspondingly in Table 4. From the table we can see that
the negative effect of family size on the likelihood of completing middle school
is monotonically decreasing by the year of birth, and the first cohort people are
the most affected group, whereas the effect on the second cohort is smaller. In
addition to the effects driven by family size, it could be possible that the
improvements in education access facilitates the educational attainment for the
later cohort; however, given that I have used the number of full time middle
school teachers at the province level as an approximation of the changes in
access to education, I have already controlled, to some extent, for such potential
effects. The results in Table 4 support the hypothesis that a second child

Table 4 Heterogeneous Effect of Family Size on Children’s Education across Cohort

DOB<1985 1985≤DOB≤1990
Independent
variable First Stage 2SLS First Stage 2SLS
(1) (2) (3) (4)
Affected* Urban –0.3252** –0.3619**
-- --
(0.1170) (0.1182)
***
Number of –0.2177 –0.1498*
-- --
siblings (0.0620) (0.0770)
Demographic Yes Yes Yes Yes
controls
Supply side Yes Yes Yes Yes
control
Province fixed Yes Yes Yes Yes
effect
Date of birth year Yes Yes Yes Yes
fixed effect
Birth order fixed Yes Yes Yes Yes
effect
Province-specific Yes Yes Yes Yes
cohort trends
N 4,117 4,117 3,669 3,669
Note: Standard errors are clustered at the province level and reported in parentheses.
*** P<0.01, ** P<0.05, *P<0.10. All control variables and fixed effects are the same as in
Table 2. The cohorts are divided by the year of birth.
56 Ying Shen

decreases the probability of completing middle school education through a

“dilution effect” channel. In the past few decades, China has experienced rapid
economic growth, going from an extremely poor country to the world’s
second-largest economy; therefore, it seems likely to be true that people born in
earlier years suffered more than those born later from any increase in number of
siblings, due to the limited parental resources.

5.2.3 Heterogeneous Effects across Birth Order

Since family size may have different effects on different children within a family,
I next turn my attention to birth order. I estimate regressions for the oldest
children and youngest children separately. Because the policy mainly generates
variation in the number of siblings between zero and one, for most urban families,
the only child is both the oldest child and the youngest child, while for many
rural families, the oldest child is the first child and the youngest child is the
second child. It is important to point out that the parents of first-borns are likely
to be younger than the parents of last-borns, thus, if parental resources change
dramatically between the years after their first child was born and before their
last child was born, the different outcomes of different children may not be fully
explained by the change in family size. In order to address this concern, I assume
that parental characteristics do not alter greatly within five years, so I include
only children whose age gap between the oldest children and youngest children is
not beyond five years. 26 The second column in Table 5 presents the second
stage result for oldest children, while the fourth column is the result for youngest
children. The table clearly suggests that the effect of family size has a larger
effect on younger children, which is consistent with Iacovou’s (2001) finding,
showing that later children have relatively lower education.

5.3 Robustness Checks

5.3.1 Sample Selection

In the baseline analysis, I exclude rural to urban migrants from consideration,

since it is difficult to determine whether households are urban residents or rural
residents, and whether they behave more like urban couples or rural couples.

26
Age gap = youngest child’s year of birth – oldest child’s year of birth.
The Effect of Family Size on Children’s Education 57

However, in this robustness check, I add them into the sample and assign their
region by Hukou.27 Specifically, if the household’s Hukou is an urban Hukou,
then I assume they are urban people, but if the household’s Hukou is a rural
Hukou, then I assume they are rural people. The first stage results and second
stage results are reported in column (1) and column (2) of Table 6, respectively.
Since all of these coefficients are quite similar to our basic regression estimates, I
am confident that excluding rural to urban migrants will not significantly bias my
results.

Table 5 Heterogeneous Effect of Family Size on Children’s Education by Birth Order

Oldest Child Youngest Child
Independent variable First Stage 2SLS First Stage 2SLS
(1) (2) (3) (4)
Affected* Urban –0.4189** -- –0.3164** --
(0.1362) (0.1286)
Number of siblings -- –0.1716** -- −0.2530**
(0.0818) (0.1009)
Demographic Yes Yes Yes Yes
controls
Supply side control Yes Yes Yes Yes
Province fixed effect Yes Yes Yes Yes
Date of birth year Yes Yes Yes Yes
fixed effect
Birth order fixed Yes Yes Yes Yes
effect
Province-specific Yes Yes Yes Yes
cohort trends
N 4,982 4,982 4,996 4,996

Note: Standard errors are clustered at the province level and reported in parentheses.
*** P<0.01, ** P<0.05, *P<0.10. All control variables and fixed effects are the same as in
Table 2. I only include the children within a family whose age gap between the oldest child
and youngest child is less than five years.

5.3.2 Other Specifications

Table 7 presents other specifications based on my identification strategy. Panel A

gives us the base results. In panel B, I add local-year level of income growth
rates to control for the income effects that the policy may address. To reduce the

27
See footnote 11.
58 Ying Shen

Table 6 Robustness Check: Including Migrants

First stage 2SLS
Independent Variable (1) (2)
Number of siblings -- –0.1898***
(0.0651)
Affected* Urban –0.3366** --
(0.1180)
Demographic controls Yes Yes
Supply side controls Yes Yes
Province fixed effect Yes Yes
Date of birth year fixed Yes Yes
effect
Birth order fixed effect Yes Yes
Province-specific cohort Yes Yes
trends
N 7,144 7,144

Note: Standard errors are clustered at the provincial level and reported in parentheses.
*** P<0.01, ** P<0.05, *P<0.10. In column (1), the dependent variable is the number of
siblings; in column (2) and (3), the dependent variable is the probability of completing middle
school. All control variables and fixed effects are the same as in Table 2. In Table 8, migrants
whose Hukou are in local urban or non-local urban areas are defined as living in an urban
place; migrants whose Hukou are in local rural or non-local rural areas are defined as living in
a rural place.

measurement error, I use the local level of annual income growth rates at an
individual’s age of 10 to 14, rather than a single year value as the control
variables. After including income growth rates at the regional level, the first stage
result loses some power, so the second stage result becomes imprecise, but it still
has a negative sign and similar magnitude. Moreover, since my identification
strategy relies on the cross-sectional variation across urban and rural, it is very
likely that urban areas have very different time trends in education than rural
areas affected by other factors other than family size, so in Panel C, I include the
urban/rural linear cohort time trends into the specification. The first stage and
second stage results are statistically insignificant, but they consistently have the
same signs as the baseline results. It is important to note that adding urban/rural
linear cohort time trends is a very demanding test that might take out some true
effects.
The Effect of Family Size on Children’s Education 59

Table 7 Robustness Check: Alternative Specifications

First Stage 2SLS
Independent variable
(1) (2)
Panel A. Base results (from Table 2 and Table 3)
Siblings –0.1786***
(0.0579)
Affected*Urban –0.3641***
(0.1168)
Panel B. Control for five years income growth
rates (aged 10–14)
Siblings –0.2330
(0.1674)
Affected*Urban –0.3072**
(0.1224)
Panel C. Control for urban/rural cohort trends
Siblings –0.1642
(0.2324)
Affected*Urban –0.1450
(0.0875)
Panel D. Cluster standard errors at the birth year
level
Siblings –0.1786*
(0.9922)
Affected*Urban 0.3641***
(0.1041)
Panel E. Cluster standard errors at four groups
level
Siblings –0.1786***
(0.0522)
Affected*Urban 0.3641***
(0.0824)
Note: Standard errors are clustered at the provincial level and reported in parentheses in the
first three panels. *** P<0.01, ** P<0.05, *P<0.10. All other control variables and fixed
effects are the same as in Table 2. In Panel D, the standard errors are clustered at the birth year
level; in Panel E, the standard errors are clustered at four groups level.

Additionally, since the F-statistics of the first stage coefficient is 10,28 some
people may worry that the results might be sensitive to the clustering methods I
choose. To alleviate such concern, in Panel D and Panel E of Table 8, rather than
clustering standard errors at the province level, I cluster the standard errors at the
birth year level and four groups level,29 respectively. We can see that the first

28
See the last column of Table 2.
29
These four groups are affected*urban, affected*rural, unaffected*urban and unaffected*
rural.
60 Ying Shen

stage coefficients are still significant at the 1% level no matter what clustering
approach I use, so the weak instrument is not of particular concern in this
analysis.

Table 8 Robustness Check: Alternative Instrumental Variable (Twins)

Dependent variable: middle school completion
Independent variable First Stage 2SLS
(1) (2)
Twins 0.6619*** –0.1664*
(0.0934) (0.0867)
Demographic controls Yes Yes
Supply side controls Yes Yes
Province fixed effect Yes Yes
Date of birth year fixed
Yes Yes
effect
Birth order fixed effect Yes Yes
Province-specific cohort
Yes Yes
trends
N 6,921 6,921
Note: Standard errors are clustered at the provincial level and reported in parentheses.
*** P<0.01, ** P<0.05, *P<0.10. The “number of siblings” is instrumented by twins. Twins
are defined as children within one family born in the same year and the same month. All
control variables and fixed effects are the same as in Table 2.

5.3.3 Alternative IV

Rosenzweig and Wolpin (1980) first used twins as an exogenous variation in

family size, and Black, Devereux and Salvanes (2005) also use twins as an
instrumental variable to analyze the quantity-quality tradeoff question. Given the
concerns for the exclusion restrictions of my IV, in order to ensure the
completeness and reliability of my conclusion, I describe results using the twins
instrument here. Specifically, I instrument family size by whether or not the
children are twins. 30 The first stage and second stage results are reported in
Table 8. The first stage is strong and the second stage coefficient is around −0.17,
implying that the base results are quite robust by using an alternative IV.

30
Twins are defined as children within one family born in the same year and the same month.
The Effect of Family Size on Children’s Education 61

6 Conclusion
In this paper I use both time-series and regional exogenous variations in family
size induced by the family planning policy in China along with a set of controls
and fixed effects, to examine the effect of family size on children’s education,
and to evaluate the effect of the family planning policy on educational attainment.
I find evidence that the compulsory fertility control policy, indeed, effectively
reduces family size in China, in particular for urban families. The relationship
between family size and child education proves to be negative, and the estimated
effect using my IV strategy is greater in magnitude than the standard OLS
strategy that does not resolve the problem of joint determination of child quality
and quantity. On average, an extra sibling leads to a relative seventeen
percentage points decline in the likelihood of completing middle school, and the
effect is heterogeneous across cohorts and the birth order of children. The
negative effect of family size is larger for the cohorts born in earlier years after
the policy, and children with the highest birth order within a family.
It is beyond the scope of this paper to look at other interesting education
outcomes. For example, data limitation did not allow an examination of the
relationship between family size and the probability of studying abroad at the
individual level; however, since the number of Chinese overseas students has
increased sharply in recent years, 31 it is a relationship worth attention when I
can get a better data set. Another issue remains unsolved: I do not provide
conclusive evidence on the mechanisms driving the main effects through “peer
effect” or “dilution effect” because the spillover effect is really difficult to
capture at the individual level. Future work is needed to explore a better strategy
to distinguish these two effects.

References
Angrist J D, Evans W N (1998). Children and their parents’ labor supply: Evidence from
exogenous variations in family size. American Economic Review, (3): 450–477
Banerjee A, Meng X, Qian N (2010). The life cycle model and household savings: Micro
evidence from urban China. Working paper
Becker G S, Lewis H G (1973). On the interaction between the quantity and quality of children.
Journal of Political Economy, 81(2): S279–S288

31
From 1978 to 2007, the total number of international students from China reached 1.2
million (China Statistical Yearbook, 2010).
62 Ying Shen

Becker G S, Tomes N (1976). Child endowments and the quantity and quality of children.
Journal of Political Economic, 84(4): S143–S162
Behrman J, Pollak R A, Taubman P (1989). Family resources, family size and access to
financing for education. Journal of Political Economy, 93(2): 389–419
Behrman J, Taubman P (1986). Birth order, schooling, and earnings. Journal of Labor
Economics, 4(3): 121–S145
Black S E, Devereux P J, Salvanes K G (2005). The more the merrier? The effect of family
size and birth order on children’s education. Quarterly Journal of Economics, 120(2):
669–700
Black S E, Devereux P J, Salvanes K G (2005). Why the apple doesn’t fall far: Understanding
intergenerational transmission of human capital. American Economic Review, 437–449
Bleakley H, Chin A (2004). Language skills and earnings: Evidence from childhood
immigrants. Review of Economics and Statistics, 86(2): 481–696
Borjas G J, Doran K B (2012). The collapse of the Soviet Union and the productivity if
American mathematics. Quarterly Journal of Economics, 127(3): 1143–1203
Donald S G, Lang K (2007). Inference with difference-in-differences and other panel data.
Review of Economics and Statistics, 89(2): 221–233
Evans W N, Schwab R M (1995). Finishing high school and starting college: Do Catholic
school make a difference? Quarterly Journal of Economics, 110: 941–974
Hanushek E A (1992). The trade-off between child quantity and quality. Journal of Political
Economy, 100: 84–117
Hungerman D M (2014). The effect of education on religion: Evidence from compulsory
education laws. Journal of Economic Behavior & Organization, 104: 52–63
Iacavou M (2001). Family composition and children’s educational outcomes. Working paper
Lafortune J, Lee S (2010). All for one? Family size and children’s educational distribution
under credit constraint. Working paper
Lee Jungmin (2008). Sibling size and investment in children’s education: an Asian instrument.
Journal of Population Economics, 21(4): 855–875
Li B (2011). The effect of family size on children quality: employing China’s one child-policy
as a natural experiment. Working paper
Li H, Zhang J, Zhu Y (2005). The effect of the one-child policy on fertility in China:
Identification based on the differences-in-differences. Working paper
Liu H (2014). The quality-quantity trade-off: Evidence from the relaxation of China’s
one-child policy, Journal of Population Economics, 27: 565–602
Lu Y, Treiman D J (2008). The effect of sibship size on educational attainment in China:
Period variations. American Sociological Review, 73(5): 813–834
Qian N (2013). Quantity-quality and the one child policy: The one child disadvantage in
school enrollment. Working paper
Rosenzweig M R, Wolpin K I (1980). Testing the quantity-quality fertility model: The use of
twins as a natural experiment. Econometrica, 48(1): 227–240
Song Y (2012). Poverty reduction in China: The contribution of popularizing primary
The Effect of Family Size on Children’s Education 63

education. China and World Economy, 20(1): 105–122

Therese H, Li L, Zhu W (2005). The effect of China’s one-child family policy after 25 years.
New England Journal of Medicine, September, 353(11): 1171–1176
Truong S A, John K, David L, Jed F (1998). Family size and children’s education in Vietnam.
Demography, 35(1): 57–70
Wu X, Li L (2012). Family size and maternal health: Evidence from the one-child policy in
China. Journal of Population Economics, 25(4): 1341–1364
Wozinak A K (2010). Are college graduates more responsive to distant labor market
opportunities? Journal of Human Resources, 45(4): 944–970
Wang D (2003). China’ rural compulsory education: Current situation, problems and policy
alternatives. Working paper
Yang J (2006). Has the One-Child policy improved adolescents’ educational wellbeing in
China? Working paper
Zhang J, Byron S (1992). Who signs China’s one-child certificate, and why? Journal of
Population Economics, 5(3): 203–215
Zhang L (2009). Disparity between rural and urban education in China, Working paper
Institute for the Study of Labor (IZA), Australian National University, University of
Queensland and Beijing Normal University (2014). Longitudinal Survey on Rural Urban
Migration in China (RUMiC) (2008). IDSC of IZA. http://dx.doi.org/
10.15185/izadp.7680.1

Appendix

Figure A1 (a) Student-Teacher Ratio in Primary School, 1996–2001

64 Ying Shen

Figure A1 (b) Student-Teacher Ratio in Middle School, 1996–2001

Source: National Statistics Bureau, China Statistics Year Book (1996–2002).

Figure A2 (a) Educational Expenditure per Student in Primary Education (yuan), 1999–2003
 The Effect of Family Size on Children’s Education 65

Figure A2 (b) Educational Expenditure per Student in Middle School (yuan), 1999–2003

Figure A2 (c) Total Education Expenditures in Compulsory Education (100 million yuan),
1996–2001
Source: Figure A2a and Figure A2b are from Yang (2007); Figure A2c are from the National
Statistical Bureau, China Educational Finance Statistical Yearbooks (1997–2002).

View publication stats