Intergenerational Mobility in China: Patterns and Determinants ∗ Yi Fan

Intergenerational Mobility in China:
Patterns and Determinants∗
Yi Fan
Junjian Yi
Junsen Zhang
May 28, 2013
Abstract
This paper estimates the time, gender, and regional patterns of intergenerational mobility in education and income in China. We find intergenerational mobility in both education and income
has been decreasing. Specifically, one year of parental schooling correlates with additional 0.33
schooling years for children born in 1975-1979, which increases to 0.37 years for children born
in 1965-1969. The intergenerational income elasticity increases from 0.32 for cohorts born before
1970 to 0.44 for cohorts born after 1970. The decreasing pattern is more evident for females, rural
residents, and residents from western provinces. We further explore mechanisms translating the
market-oriented institutional reforms and socioeconomic changes to our estimated patterns under
a unified economic model. The decrease in intergenerational mobility is driven by the increases in
both return to human capital and educational costs. The gender disparity is due to gender difference
in return to human capital. The regional disparity is caused by the localization of governmental
budget on education. The intergenerational estimates suggest that cross-sectional inequality will
deteriorate for the next generation in China.
∗
Yi Fan: Department of International Development, London School of Economics and Political Science; email:
[email protected] Junjian Yi, Department of Economics, University of Chicago; email: [email protected]
Junsen Zhang, Department of Economics, Chinese University of Hong Kong; email: [email protected]
1
Introduction
China has experienced remarkable economic growth since its market-oriented reform in 1978
(Brandt and Rawski, 2008). Along with its GDP growth, the rising inequality is also striking.
The Gini coefficient rockets from 0.26 to 0.43 between 1980 and 2010, equal to the US level (Figures 1 and 2). The rural-urban difference and regional disparity are enlarged as well (Figures 3
and 4). Literature has investigated the cross-sectional inequality (Li et al., 2005; Fleisher et al.,
2010; Li et al., 2012), though not much attention has been paid to the intergenerational inequality.
However, the patterns and mechanisms of intergenerational mobiltiy is crucial, as they estimate
whether and how the intra-generational inequality can be passed down to the next generation, and
predict the dynamic of the society. In this paper, we examine the intergenerational mobility in both
education and income in the context of China’s market reform, and theorize the determiants in the
transmission across generations.
Based on Becker and Tomes (1979, 1986)’s cornerstone paper crystallizing intergenerational
mobility, and Solon (2004) rationalizing the commonly used log-linear regression, we innovatively
introduce the share of households subject to credit constraint into the classical model. We characterize determinants of intergenerational mobility along with China’s market-oriented reforms.
They are return to human capital, tightness of credit constraint (which is determined by progressivity of government expenditure on public education and price of human capital investment relative
to consumption goods), and the share of parents subject to credit constraint.
In addition, we present empirial evidence on the decreasing intergenerational mobility in both
education and income. Applying the Chinese Family Panel Studies in 2010, we find that with
one more year of parental average schooling, children’s schooling is increased by 0.374 among
1975-1979 cohort. It is 0.047 larger in magnitude than the one in 1965-1969 cohort, and with statistical significance at 5% level. Ajusted for different dispersion across generations, the correlation
reveals similar pattern. Using Chinese Household Income Projects in 1995 and 2002, we evidence
an increase from 0.315 to 0.442 in intergenerational elasticity (IGE) in income between cohorts
born before and after 1970. The differences in IGE and correlation across cohorts are statistically
1
significant at conventional levels of significance. The decreasing intergenerational mobility in education and income is driven mainly by the female, and residents from economically disadvantaged
regions, such as rural and western parts.
We link determinants derived from our model to interpret the estimated decreasing mobility
across generations. Evident rising return to human capital and increasing educational costs lead
to decrease in intergenerational mobility (data source of figure 6?; NBS, 2013). Expanded government expenditure on public education and fast economic growth which relaxes households’
credit constraint raise the intergenerational mobility (NBS, 2012 and 2013). According to our
empirical evidence, we propose that the former force dominates the latter along China’s market
reform. Thus an overall decreasing pattern in intergenerational mobility is revealed. We attribute
the gender-specific pattern to the interaction between residual son preference and higher increasing return to human capital for girls. The regional disparity is ascribed to inequality in localized
government expenditure on public education and differential income levels across regions.
Early empirical paper on intergenerational income mobility focuses on point estimation between US fathers and sons, with a low estimate of 0.2 (Sewell and Hauser, 1975; Behrman and
Taubman, 1985). Latter researchers concentrate on improving the measurement of intergenerational elasticity (Solon, 1989a, 1992), and examining the trend (Levine and Mazumder, 2002;
Mayer and Lopoo, 2005; Bratsberg et al., 2007; Lee and Solon, 2009). A larger estimate of 0.4
for US fathers and sons is yielded, with increasing or non-changing trend in the IGE (Mayer and
Lopoo, 2005; Lee and Solon, 2009).
However, little is known about the intergenerational income transmission in developing countries as China. Pioneer work on estimating IGE between Chinese fathers and sons varies between
0.32 and 0.74 (Labar, 2007; Guo and Min, 2008; Gong et al., 2010; Deng et al., 2011). We evidence decreasing intergenerational mobility across time, gender and region, revealing a similar
pattern as that in Chen et al. (2010) that the more open the economic institutions are the less the
social mobility.
Beyond traditional transmitting mechanism of education, recent research explores new inter-
2
generational transmitting mechanisms such as non-cognitive skills (Blanden et al., 2007), health
(Eriksson et al., 2005; Currie and Moretti, 2007), personality (Groves, 2005), and political participation (Ichino et al., 2011). Nevertheless, we focus on the classical channel of education in this
paper (Bowles and Gintis, 2002; Blanden et al., 2004). Our aim is to analyse in detail which and
how factors affecting human capital investment along with China’s market-oriented reforms. We
concentrate on the effects of expansion of higher education, increasing tuition fees, decentralization of government expenditure on education, and family credit constraint.
Because education is the only source generating earnings in our model, our theoretical framework can also be applied to analyze intergenerational educational mobility. Estimation on schooling has advantages over income as the measurement error is less a problem (Black et al., 2011).
Literature evidences significant correlation between parental and children’s education, but mainly
from developed countries (Bowles, 1972; Couch and Dunn, 1997; Blanden and Machin, 2004;
Bauer and Riphahn, 2009). Hertz et al. (2007) survey 42 countries across 50 years, ranking the
international educational mobility. They report estimated regression coefficients of 0.46 and 0.34
for US and rural China, respectively.
Not much work has been done on the intergenerational schooling mobility in China. Knight
and Li (1993) pioneer in showing that parental education is more important for daughters than
sons in rural China. The regional difference is significant in affecting individuals’ educational
attainment. Knight et al. (2010) estimate the intergenerational schooling coefficients as 0.361
for sons, 0.448 for daughters, 0.416 for rural residents, and 0.402 for urban ones. Our estimates
are roughly the same as theirs. We further extend the empirical research to examine trends with
time and across geographic locations, and incorporate structural determiants into the framework of
interpreting changing patterns.
Comparing the estimates between US and China, we find the point estimates are almost the
same, 0.44 (Lee and Solon, 2009) versus 0.442 (Table 5) in intergenerational income elasticity,
and 0.46 (Hertz et al., 2007) versus 0.374 (Table 2) in educational coefficients. However, the
trends differ from each other. Unlike the unchanging or increasing trends found in US (Mayer
3
and Lopoo, 2005; Lee and Solon, 2009), we evidence a statistically significant decreasing trend
in China’s intergenerational mobility. We ascribe it to the rise in educational costs along with the
market reform, which reinforces the increasing return to human capital, and outweighs the effect
of expansion of government educational expenditure on intergenerational mobility. Scenario in
the US is on the opposite. Rise in federal and state government expenditure on public education
possibly dominates the impact from increasing return to human capital, leading to a non-changing
or increasing intergenerational mobility.
This paper contributes to the literature in three ways. First, we innovatively introduce share of
credit-constraint households and price of human capital relative to consumption goods into classical model, structuring four main determinants for intergenerational mobility along with China’s
institutional reforms. Second, we evidence China’s decreasing intergenerational mobility in education and income by time, region, and gender, incorporating factors derived from mode to interpret
the estimated patterns. Last but not least, we compare the intergenerational mobility between
China and US, articulating the differences in patterns lying in the bargaining between progressivity of government expenditure on education and increasing return to human capital. To our
best knowledge, this is the first paper systematically theoretizes and evidences China’s decreasing
intergenerational mobility in education and income along with the market reforms, and structure
differences in mechanisms between US and China. According to the model and empirics, we expect the future cross-sectional inequality would rise if the households’ credit constraint were not
relaxed effectively.
The rest of the paper is organized as follows. Section 2 provides background information
on China’s market-oriented institutional reforms, along with its educational and fiscal policy reforms. Section 3 sets up a unified model theorizing mechanisms of intergenerational mobility, and
deriving determinants of mobility across generations. Section 4 presents empirical results of intergenerational mobility in education and income, followed by section 5 interpreting changing trends
and patterns. Section 6 discusses and concludes.
4
2
Research Background
This section describes research background. We first introduce key market-oriented institutional
reforms in China. We then discuss educational reforms and primary and secondary education in
China. Finally, we focus on the tertiary education expansion in China since 1999. Fiscal policy
reforms relating to education are also discussed in detail.
2.1
Market-oriented Institutional Reforms
China has experienced unprecedent economic growth during the past four decades. China was
one of the poorest country in the world in 1978. The real per capita GDP in China was only onefortieth of that of US before the economic reform in 1979. Since then, China has experienced
fundamentally structural reforms and the annual growth rate of per capita GDP has exceeded 8%.
In 2012, the real per capita GDP in China was one-fifth of that of US.
The China economic growth is promoted by a series of market-oriented institutional reforms
(Zhu, 2012). The economic reform was initiated by the implementation of the rural household
responsibility system in late 1970s and early 1980s (Lin, 1988). It adjusted the incentive structure
in the agricultural sector, enhanced substantially the agricultural productivity, and led to a surplus
of labor in rural areas. During the same time, the Chinese government adopted an Open-Door
policy which induced a huge influx of foreign direct investment (FDI). The large inflow of physical
capital created a strong demand for labor in urban and coastal areas.
Facing both the surplus of labor in rural and inland areas and the strong demand of labor in
urban and coastal areas, China gradually relax the strict hukou system in late 1980s and early
1990s. Labor mobility between rural and urban areas is virtually not allowed before the economic
reform in 1979. A household registration system called hukou segregates China into two labor
markets. Individuals born in rural areas received “agricultural hukou” and those born in urban
areas received “non-agricultural hukou.” The relaxation of the hukou system led to a historically
unprecedent domestic migration. The total rural-to-urban migrants amounts to 0.2 billion based on
5
the 2005 1% mini-census. The allocative efficiency of the labor market has improved greatly by
the relaxation of the hukou system.
In late 1990s, the government began to encourage the diversification of ownership structure
and promote the reforms of state owned enterprises (SOEs). The 15th Congress of the Chinese
Communist Party in 1997 formally sanctioned ownership reforms of the SOEs and legalized the
private enterprises. Before the economic reform, the stated-owned and collective enterprises are
the only two types of firms. With the Open-Door policy, the three types of foreign enterprises
-Chinese-foreign joint venture management enterprise, Chinese-foreign cooperative joint venture,
and Foreign sole-source investment enterprise- were allowed. With the restrictions on private
firms, the collective enterprises such as township and village enterprises were burgeoning in early
and middle 1990s. The formal legalization of the private firms make collective enterprise lose their
edge. Many of them are privatized. During the same time, the government began to privatize SOEs
in large scales. The market-oriented privatization process results in a more flexible labor market
and promote economic growth.
All these key market-oriented institutional reforms adjust the incentive structure, enhance the
labor productivity, and increased private return to human capital. Li et al. (2012) review the change
in wage structure and in return to education in China since the economic reform. Figure (5) shows
that gap in annual wage between the low-education and medium-education groups is almost zero
in 1988. In the same year, the annual wage of the high-education group was only marginally higher
than the other two groups. In contrast, the annual wage of the high-education group was as twice as
that of the low-education group, and 1.5 times as that of the medium-education group. This pattern
is confirmed in Figure (6). It shows that the return to one additional year of schooling increases
five times from 2% in 1998 to 10% in 2008. The increase in return to college education vs. high
school is more dramatic, from 7% to 49% during the same period.
The increase in return to education stimulates both the family and the government to investment
on children’s education. We then discuss Chinese education in the past decades.
6
2.2
Primary and Secondary Education
Besides the market-oriented institutional reforms, the Chinese economic growth benefits from the
large stock of medium skilled labor acquired before the economic reform (Heckman and Yi, 2012).
Figure 7 shows that the primary school (grades 1-6) enrollment rate was 98% and the junior secondary school progression rates, that is, the ratio of junior secondary school (grades 7-9) enrollments over the primary school graduates, is 70% in 1981. Becker et al. (2010) shows that the
primary and junior secondary completion rates were much higher in China than that in other Asian
countries, such as South Korea and Taiwan in the early stage of their economic take-off after adjusting for stages of economic development.
With the economic reform, the secondary school progression rates have risen substantially in
the past four decades. Figure 7 shows that the junior secondary school progression rate began to
rise in 1990. It has been at almost 100% since 2000. During the same period, the senior secondary
school (grades 10-12) progression rates have also experienced substantial increase. It rose from
26% in 1981 to 82% in 2008.
Despite the substantial progress made in primary and secondary education, educational investments have been made by households and local governments until recently. Before the economic
reform, the education costs were borne collectively. With the economic reform, these costs were
shifts to households, especially for those in rural areas.
The fiscal policies which are relating to finance primary and secondary education have also
experiences several changes. The government adopted fiscal decentralization in the mid-1980s,
which developed multiple sources of funding for education by local government. In urban areas,
district governments were responsible for the primary schools, and the city governments were
responsible for secondary schools. In rural areas, county, township, and village were responsible
for senior secondary, junior secondary, and primary schools, respectively. Under this policy, the
local government heavily relied on surtaxes, tuition, and other fees to finance education. Education
funding was localized. The access to education and the quality of education became increasingly
unequal.
7
A tax reform and fiscal re-centralization were carried in 1994 which strengthened the central
government’s fiscal capacity. Under this reform, the central government transfer partly financed
the primary and secondary education. local governmental funding had to matching the government
transfer to finance the other part for the primary and secondary education. This reform exacerbated
the regional inequality in the public finance. The central-local transfers were insufficient. Local
governments, especially those in poor rural areas were unable to meet their obligations.
The public finance in primary and secondary education has substantially improved since early
2000s. The payment of teacher’s salaries was shifted from the village to the county government
in 2001. During the same year, the central government initiated the “Two Exemptions and One
Subsidies” policy. Under this program, the government aimed to pay tuition and other fees and
textbook costs and subsidy boarding for all primary and junior secondary school students. The
objective of this program was achieved in 2006 when all tuition and fees for rural compulsory
nine-year education were exempted. But the senior secondary school is still not covered by this
program.
2.3
Tertiary Education
The Chinese tertiary education have experienced fundamental change since 1999. The tertiary
education was suspended during the Cultural Revolution from 1966 to 1976. Tertiary school enrollment rates hovered around 1.5% to 2.5% from 1979 to 1995 (NBS, 2011a). With the economic
growth and fast accumulation of physical capital, the return to education increased (Wang, 2012).
China maintained a national gross saving rate as high as 35-55% throughout the entire economic
reform period, dramatically increasing the marginal productivity of labor, especially highly skilled
labor. The newer technology was strongly complementary with higher skilled labor). The increase
in the demand for higher education led the government to implement a radical policy of expansion
of higher education in 1999. The total number of fresh college graduates increased more than sixfold from less than one million in 2001 to seven million in 2013 (NBS, 2011a). Figure 8 displays
a jump of the share of college students in the age cohort 18-22.
8
With the higher education expansion, the private costs of tertiary education increased drastically. Higher education was heavily subsidized before late 1990s. The annual tuition increased
from RMB 800 in 1995 to RMB 5,000 in 2004 per person (li and Xing, 2010). Based on a national
survey on college student conducted by Tsinghua University in 2010, the annual expenditure per
college students was RMB 12,318 (Li, 2013). If we define a poverty line that the totaly family
income of a college student is less than RMB 12,318, the poverty rate is 22% for all college students in 2010. Li (2013) further finds that 76% of the expenditure of college students comes from
their parents. The college student loans and scholarship account for about 10% of the expenditure.
Furthermore, more than one third of the students have to borrow money to finance their college
education. The drastic increase in the costs greatly exacerbate the inequality in access to higher
education, especially for those from poor and rural areas.
The higher education expansion had accompanied with the decentralization of the administration of colleges and universities during the 1990s. The higher education system developed into a
two-layer system. The central government administrates a small quantity of distinguished universities. The local governments at different levels administrate most local colleges and universities.
Therefore, the regional disparity in higher education resources was amplified by the higher education expansion.
2.4
Public and Private Investments in Education
To get a better understanding of the public finance in education, Figure 9 depicts the share of
government expenditure on education out of GDP from 1992 to 2012. It shows that the share. We
find the ratio have almost doubled during this time period. It increases from 2.4% in 1992 to 4.4%
in 2012. Comparing with the marvelous achievements in both secondary and tertiary education
described above, the increase in the share of public expenditure over GDP seems to be mild. Then
the question is who finance the substantial improvement in education in China. The answer is the
household.
The increase in private education costs has been more dramatically than the increase in public
9
expenditure on child human capital. Figure 10 shows the time trend of tuition relative to governmental expenditure on education and the GDP from 1991 to 2007. The ratio of tuition over
government expenditure on education increased seven times from 5% in 1991 to 35% in 2004, followed by a mild decrease after the government initiated the “Two Exemptions and One Subsidy”
program. The ratio of tuition over GDP increased eight times from 0.1% in 1991 to 0.8% in 2007.
This fact has severely exacerbated the family credit constraint in invest the child human capital and
then led to inequality in access to education.
To understand how these market-oriented institutional, educational reforms, and fiscal reforms
affect intergenerational mobility in China, we need a theoretical model which is presented in the
section below.
3
Model
This section formulate an intergenerational mobility model which is based on Becker and Tomes
(1986) and Solon (2004). We incorporate the institutional reforms and socioeconomic changes
into a unified theoretical framework, and use it to explain our estimated time,regional, and gender
patterns in China in the past few decades. This model also help us understand the difference
in intergenerational mobility between China and US, and the cross-sectional inequality for next
generation in China.
3.1
Model Setup
We assume for simplicity that family i contains one parent of generation t − 1 and one child of
generation t.1 The parents are altruistic and care about their own consumption (ci,t−1 ) and the
earnings of their children (yt ). Thus, parental preference can be represented by a utility function of
1
In our empirical analysis, we use the average parental schooling years or the total parental incomes as dependent
variable. Therefore, the issue of assortative mating has been well taken care of in our analysis, and it is suitable to
assume that there is only one parent in the family.
10
the following form:
Ui,t−1 = (1 − a) ln Ci,t−1 + α ln yt ,
(1)
where 0 < α < 1 measuring the degree of parental altruism toward their child. The parents allocate
their after-tax earnings ((1 − τ)yi,t−1 ( between the parents’ own consumption (Ci,t−1 ) and investment
in the child’s human capital (Ii,t1 ). Thus, the budget constraint is
(1 − τt )yi,t−1 = Ci,t−1 + pt−1 Ii,t−1 ,
(2)
where τ is the tax rate. We normalize the price of consumption good to one and pt−1 is the price of
human capital investment. Different from the literature, we use pt−1 to capture the rapid increase
in costs of human capital investment such as tuition in China in the past decades.
The child’s income is generated from the following semi-log function
ln yit = µ + rt hit ,
(3)
where hit is the level of human capital and rt is the monetary return to human capital. We note that
var(ln yit ) = rt2 var(hit ).
Given the distribution of human capital stock, the higher the return to human capital, the greater the
cross-sectional inequality. Our theoretical model explore the effect of the return to human capital
on the intergenerational mobility and the cross-sectional inequality for the next generation.
The return to human capital is determined by three factors: the stock of physical capital (Kt ),
technology (At ), and the market-oriented institutional reforms (Mt ) such that
rt = r(Kt , At , Mt ).
The stock of physical capital and the technologic progress enhance the marginal productivity of
11
human capital, and then increase rt . Thus, we have ∂rt /∂Kt > 0 and ∂rt /∂At > 0. Furthermore, the
market-oriented instructional reforms not only increase the marginal productivity of human capital
but also shrink the gap between the marginal productivity and the market wage. Thus, we have
∂rt /∂Mt > 0. The institutional reforms play a big role in the return to human capital in developing
countries such as China.
The human capital is produced by the following equation
hit = ln(Ii,t−1 + Gi,t−1 ) + eit ,
(4)
where Gi,t−1 is the governmental investment in child human capital, and eit is the child endowment.
We assume that governmental and family investments are substitutes. Finally, we assume that the
endowment transmission follows a first-order autoregressive process such that
eit = δ + λt ei,t−1 + vit ,
(5)
where vit is i.i.d.
3.2
Intergenerational Mobility without Credit Constraints
We first consider the case that the credit market is perfect that parents can borrow against the
child’s prospective earnings or parents are sufficient rich to leave positive bequest to the child. In
this case, the child human capital investment decision is independent of parental income. It is easy
to derive a log-linear intergenerational income regression
ln yit = µ1t + λt gt ln yi,t−1 + v1it
where µ1t is a constant and v1it is uncorrelated with ln yi,t−1 . gt = rt /rt−1 which measures the growth
rate of return to human capital between the two generations.
In this case, the intergenerational transmission of endowments is the only mechanism bridging
12
the parental and child’s earnings. The growth of return to human capital can inflate the effect of this
mechanism on intergenerational mobility. The higher the increase in return to human capital, the
lower the intergenerational mobility. When it is at the steady state that rt = rt−1 and λt = λt−1 , the
intergenerational earning elasticity is uniquely determined by the intergenerational transmission of
endowment such that
β1 = λ
3.3
Intergenerational Mobility with Credit Constraints
Parents generally cannot borrow against the child’s future earnings and the great major of households is not sufficient rich. We thus consider the case of intergenerational mobility with credit
constraints. In this case, the child human capital investment decision is determined by parental
income. By some simple algebras, the log-linear intergenerational income regression equation is
derived as follows
ln yit ≈ µ2t + [(1 − γt )rt ] ln yi,t−1 + eit ,
(6)
where µ2t is a constant.2 γt is function of two variables. The first one is the progressivity in government investment in child human capital (st ) measured by the share of government expenditure
out of household disposable income st = Gi,t−1 /[(1 − τt )yi,t−1 ]. The second variable is the price of
human capital investment (pt−1 ). Thus,
γt = γ(st1 , pt ).
We have ∂(1 − γt )/∂st < 0 and ∂(1 − γt )/∂pt > 0. Therefore, we interpret 1 − γt as the severity
of credit constraint. When the share of government expenditure is higher or the price of human
capital investment is lower, the credit constraint is less severe.
Equation 6 shows two channels through which that parental income affects the child’s income.
The first one is relating to the credit constraint ((1 − γt )rt ). The severe the credit constraint, the
2
We use an approximate equality because we derive the expression by using a first-order Taylor approximation.
13
greater the intergenerational elasticity of income. The increase in government expenditure on
child human capital relative to household disposable income promote intergenerational mobility,
whereas the increase in price of human capital investment decrease intergenerational moblity. Furthermore, the effect of the severity in credit constraint is inflated by the return to human capital
in the labor market. The second channel is again through the intergenerational transmission of
endowments because the last term in Equation 6 is a function is parental endowment which is
determinant of parental income.
We then derive the intergenerational income elasticity at steady state. Equation 6 is a firstorder autoregression of ln yit with a serially correlated error term that itself follows a first-order
autoregression. As shown in Solon (2004), the steady-state intergenerational income elasticity is
β2 =
(1 − γ)r + λ
.
1 + (1 − γ)rλ
(7)
We find that ∂β2 /∂r > 0, ∂β2 /∂(1 − γ) > 0, ∂β2 /∂(1 − λ) > 0. Furthermore, β2 > β1 .
3.4
Summary of Theoretical Determinants of Intergenerational Mobility
In reality, some households are subject to credit constraints to child human capital investment and
other households are not. The estimated intergenerational income elasticity should always be a
combination of the two cases. Therefore, the intergenerational income elasticity for a population
with two types of households is as follows
β = (1 − π)β1 + πβ2 ,
where π =
2
i=1 σln yit ln yit−1
PN 2
i=1 σln yit−1
Pn
and Nn (= d) is the share of parents who are subject to credit constraint. The
intergenerational elasticity of income is affected by the share of the households subject to credit
constraint in a population. Specifically, the more the households subject to credit constraint, the
higher the intergenerational elasticity of income (∂βt /∂dt > 0). The reason is that β1 < β2 and
∂πt /∂dt > 0.
14
The share of parents who are subject to credit constraint is determined by the equilibrium
condition that the marginal return to human capital investment equals to interest rate in the finance
market. When the return to human capital investment is higher than interest rate, parents are
subject to credit constraint and they do not leave positive bequest to their children. Otherwise,
parents invest the child’s human capital at the point where the marginal return to human capital
investment equals to interest rate and leave positive bequest to their child. Therefore, d is function
of the following form
d = d(Y, r, s, p, ra ),
where ra is the return to asset or interest rate. We have ∂d/∂Y < 0, ∂d/∂s < 0, ∂d/∂ra < 0,
∂d/∂r > 0, and ∂d/∂p > 0. Therefore, the government expenditure on child human capital and the
price of human capital investment do not only have an intensively marginal but also an extensively
marginal effect on the intergenerational income elasticity. The former refers to the change in the
severity of credit constraint, and the latter refers to the change in the share of households subject
to credit constraint.
We summarize theoretical determinants of intergenerational income elasticity in our model.
The market-oriented institutional reforms and socioeconomic changes affect the intergenerational
mobility functioning exclusively through these theoretical variables. If we assume that the intergenerational transmission of endowment is constant, there are three categorise of determinants.
• The return to human capital (r). It is determined by the stock of physical capital (K), technology progress (A), and the degree of marketerization which is induced by the market-oriented
reforms (M). The return to human capital inflates the effect of credit constraint of child
human capital investment on intergenerational income elasticity.
• The severity of credit constraint (1 − γ). Given household income, it is determined by the
progressivity of government expenditure on child human capital (s) and the price of human
capital investment (p). The severity of credit constraint can be interpreted as an intensive
margin of the effect of credit constraint on intergenerational income elasticity.
15
• The share of parents who are subject to credit constraint (d). Given return to human capital, the progressivity of progressivity of government expenditure on child human capital,
the price of human capital investment, and the market interest rate, it is determined by the
household income (Y). The share of parents who are subject to credit constraint can be interpreted as an extensive margin of the effect of credit constraint on intergenerational income
elasticity.
The next section presents the empirical patterns of intergenerational mobility by time, gender,
and region in China in past decades. We then use this theoretical framework combining with the
institutional and socioeconomic changes to explain the estimated patterns. Finally, we note that
the theoretical model outlined in this section focuses on intergenerational mobility in income. In
our empirical analysis, we study both intergenerational educational and income mobility. Because
human capital is the only source in the earning generating equation, the theoretical framework can
also be applied to analyze intergenerational educational mobility.
4
Empirical Results
In this section, we present empirical findings on intergenerational mobility in education and income
in recent decades of China. We evidence decreasing trend in intergenerational mobility along with
the market-oriented reform. By decomposing intergenerational mobility into various categories,
we find this decrease in mobility is driven mainly by the female, the rural and western residents.
4.1
4.1.1
Intergenerational Mobility in Education
The Chinese Family Panel Studies
We apply the Chinese Family Panel Studies in 2010 to investigate the intergenerational educational
mobility. The Chinese Family Panel Studies aim at collecting and tracking individual-, family-, and
community- level data, to investigate the socioeconomic and demographic changes in China. The
survey is funded by Peking Universtity, and sponsored by the Ministry of Education and National
16
Natural Science Foundation of China. The 2010 survey is implemented by the Institute of Social
Science Survey at Peking University. It covers 25 provinces, municipalities, and autonomous
regions, including approximately 15,000 households.
Merits of using this data set to examine intergenerational mobility in education lies in the following three aspects. First of all, it collects information on individuals both are present or not living
at home during the enumeration. Thus it overcomes the coresiding bias in conventional household
survey, and is representative for general population. Second, it classifies schooling levels into
detailed categories between illiteracy and doctorial degree, which facilitates our calculation of accurate schooling years. Last, it records clear household-member relation and basic demographic
and socioeconomic information such as Hukou status, family income and Communist Party membership. We therefore can identify the parent-child pairs regardless of coresidency, and analyse
intergenerational schooling correlation under different sets of controls.
Table 1 displays the summary statistics across three birth cohorts: 1965-1969, 1970-1974,
1975-1979. Children in the first cohort started education before the market reform. Those in the
second cohort obtain education at the beginning of reform.3 Children born between 1975 and 1979
were educated in the latter stage of reform. Taking assortative mating into concern, we present
parental average schooling years, along with children’s schooling. Education in both generations
increases gradually. Average schooling years for children rise from 7.14 in 1965-1969 cohort
to 8.23 among 1975-1979 cohort, approaching the completion of junior high school. Average
schooling between parents is merely 2.79 for the earliest cohort, but approaches completion of
primary school in the latest cohort (3.82 years). Child’s gender is balanced around 0.5 in our
sample, with slight bias toward girls. About 70% of the parent-child pairs are from rural areas,
which is representative for the general population in China.
Columns (2) - (6) present data description by region. Disparity in education between urban and
rural areas persists in both children’s and parental generations. Schooling years of urban children
3
The reason for choosing 1970 as a cut-off point is as follows. The economic reform started from 1978. The
normal age of joining primary school is seven in China. Thus those born after 1970 are considered to have received
education in the new era. Distinguishing the two cohorts by education is because that in the literature education is
considered as the main reason for intergenerational income persistence (Blanden et al., 2007; Chen et al., 2010).
17
almost doubles that of their rural counterparts across the three cohorts. Parental average schooling
years even more than doubled in urban China than that in rural areas. Increase in disparity between
eastern and western regions in both generations is more evident. Children’s averge schooling years
in western area raise by 6% (5.962 versus 6.302) from 1965-1969 cohort to 1975-1979 cohort. The
corresponding rise of education in eastern region rockets to 22% (7.543 versus 9.212).
4.1.2
The Econometric Specification
We regress child’s schooling years on parental average schooling years, following the literature
(Hertz et al., 2007; Knight et al., 2010). We introduce cohort dummies Cik (k = 2, 3) and examine
the trend in intergenerational educational mobility. The regression is specified as follows:
sit = α0 + α1 si,t−1 +
3
X
αkCik si,t−1 + Xi γ1 + Xi
k=2
3
X
Cik γk + it
(8)
k=2
where sit is the schooling years of children. si,t−1 is the average schooling years between fathers
and mothers. Xit is a vector of control variables, including child’s age, parental average age, child’s
gender dummy, and wave dummy. Standard errors are clustered by households, as there can be
heterogeneity across different families. We investigate intergenerational educational mobility in
three cohorts: 1965-1969, 1970-1974 (Ci2 = 1), and 1975-1979 (Ci3 = 1).
αi (i = 1, 2, 3) are what we are interested in. α1 is the regression coefficient for the 1965-1969
birth cohort. α2 and α3 are the incremental of schooling coefficients in the latter two cohorts,
respectively, compared to the earliest cohort. Thus α1 + α2 and α1 + α3 summarize the impacts of
parental schooling on children’s schooling in 1970-1974 and 1975-1979 birth cohorts, separately.
Considering the differential variance of schooling years across parental and children’s generations, we examine the intergenerational educational correlation (Hertz et al., 2007). It is defined
as follows:
intergenerational educational correlation = intergenerational educational coefficient ∗
18
σt−1
σt
(9)
where σt−1 and σt are the standard deviation of parental and children’s schooling years,
respectively.
4.1.3
The Estimation Results
Table 2 displays the regression coefficients and intergenerational schooling correlation in each
birth cohort. The first three columns show the intergenerational schooling coefficients across three
birth cohorts from Eq. (2). Columns (4) - (6) present intergenerational educational correlation,
taking the different variance across children’s and parental generations into account. The last four
columns summarise changes in coefficients and correlation between first and second, first and
third cohorts. We also display gender-specific estimates to examine differential effects of parental
schooling on boys versus daughters.
We find statistically significant increase in intergenerational coefficients and correlation in education across three birth cohorts. This rise is driven mainly by the female. One year of parental
schooling leads to additional 0.33 schooling years for children born in 1965-1969, which is 0.37
years for children born in 1975-1979. Both of the two estimates are statistically significant at a
high 1% level. The increasing trend persists in educational correlation (columns (4) - (6)), though
smaller in magnitude due to less variance of schooling in parental generation than that in children’s
generation. We find no statistically significant difference in intergenerational mobility between the
first two cohorts. However, an increase around 0.05 in both coefficients and correlation is estimated
from 1965-1969 to 1975-1979 cohorts, with statistical significance at 5% level. It evidences significant reduction in intergenerational schooling mobility along with the market-oriented reform.
In addition, we estimate lower intergenerational mobility in education for daughters than sons.
The estimated correlation for girls is larger than that of boys by 0.033-0.073 across three cohorts
(columns (4) - (6)). Moreover, we evidence a statistically significantly decreasing trend in intergenerational correlation for daughters, by 0.051 from earliest to latest cohorts (column (10)). No
such trend is revealed for sons. We interpret it as the outcome of interaction between residual
son preference and rising educational cost. The family credit constraint of investing in daughters’
19
human capital becomes tighter along with the market reform.
Table 3 shows regional disparity in intergenerational schooling mobility. We find statistically
significant rise in the correlation of education across generations in rural and western areas. No
such increasing trend is demonstrated in urban areas, eastern or central regions. Although the estimated rural correlation is smaller in each cohort than their urban counterparts (columns (4) - (6)),
it displays a statistically significant rise of 0.054 from the earliest to latest cohorts (column (10)).
In western region, the intergenerational schooling correlation is as low as 0.257 among 1965-1969
cohort, but reaches 0.338 for 1975-1979 cohort, exceeding the corresponding level in eastern area
(0.285). The significant decrease in intergenerational educational mobility in economically disadvantaged areas (rural and western parts) reveals tighter household budget constraint in investing
in children’s human capital. It results from decentralization of government expenditure on education, and sharp rise in the educational cost during the market reform. We will provide detailed
discussion in section 5.
4.2
4.2.1
Intergenerational Mobility in Income
The Chinese Household Income Project
We use Chinese Household Income Projects (CHIPs) in 1995 and 2002 to investigate the intergenerational income mobility. It is a series of annual micro-level surveys which aim at measuring
individual and household income, as well as other economic factors in China. It is a joint research
sponsored by the Institute of Economics at the Chinese Academy of Sciences, the Asian Development Bank, the Ford Foundation, and the East Asian Institute at the Columbia University. The
survey is based on face-to-face interview, and covers 1/3 of the 34 province-level administrative
units in China.4 We focus on urban areas only. Rural residents and urban-to-urban migrants are
not included.5
We consider the following advantages of using CHIPs for this research. First, it records each
4
Detailed description of the 1995 and 2002 household surveys is provided in Li et al (2008).
Migrants from rural to urban areas still hold rural registration (Hukou), and do not have equal access to educational and occupational opportunities as urban citizens.
5
20
individual’s detailed income from wage, subsidy, bonus, private business, and capital income in
the preceding six (1995 survey) or five (2002 survey) years. It provides a rare chance to calculate
lifetime income in a developing country as China. Taking average across years gets rid of random income shock in one specific year. Second, this survey records detailed relationship among
household members, which facilitates our identification on parents and children. Last but not least,
the areas under this survey are geographically and economically representative, which provides an
opportunity to yield nationally representative estimates.6
Table 4 presents the data description. The early birth cohort refers to those born between 1949
when the People’s Republic of China was founded and 1970 (included). Most of those children
complete their education before the economic reform. Children born after 1970 belongs to the late
birth cohort. Most of them obtain education and start working in the post-reform era.7 To avoid
measurement error in income from individuals who just enter the labor market, we restrict children
to be at least 23 years old, and have worked for at least three years. The average age of children is
29.52 and 25.46 in early and late cohorts, respectively, which are at the early-middle stage of life
cycle for working population. The average father’s age is 57.14 and 53.28 for each cohort, which
are at slightly late stage of working population.
To even income shock in specific year(s), we restrict those fathers to have at least three years’
recorded income. There is no restriction on mothers’ income, nevertheless, since there can be
housewives with no income especially in the early birth cohort. The household annual income
from parental is 9331.36 Yuan among early cohort, adjusted by Consumer Price Index (CPI) to the
price in 2002.8 It almost doubles to 15432.46 Yuan in the late cohort. Children’s average income
is 6628 Yuan and 8939.76 Yuan in each cohort, respectively. Statistics summarized by region is
6
CHIP is considered geographically representative since the areas under survey cover the Northeast (Liaoning), the
South (Guangdong), the Southwest (Yunnan), and the West (Gansu). It is considered to be economically representative
as the surveyed areas include the richest parts in China such as Beijing and Guangdong, as well as the least developed
parts such as Gansu.
7
The reason for choosing 1970 as a cut-off point is as follows. The economic reform started from 1978. The
normal age of joining primary school is seven in China. Thus those born after 1970 are considered to have received
education in the new era. Distinguishing the two cohorts by education is because that in the literature education is
considered as the main reason for intergenerational income persistence (Blanden et al., 2007; Chen et al., 2010).
8
9331.36 Yuan approximately equals 1521.40 USD using exchange rate on May 26, 2013.
21
also presented. East (coastal) areas have the highest income level, followed by the West and the
Central areas.
A caveat of CHIPs data is its potential bias due to coresidency. The household members are
defined as those who live together stably or temporarily do not live together but have close economic relationship.9 In this case, those children who neither live together nor have close economic
relationship with their parents are not included. This is possibly the reason why child’s gender
biases toward boys in both cohorts, as married daughters are excluded by this criterion.
4.2.2
The Econometric Specification
Regression estimating the intergenerational income elasticity follows conventional specification in
the literature (Solon 1992; Levine and Mazumder, 2002; Mayer and Lopoo, 2005; Lee and Solon,
2009). To capture the effect of institutional change on the IGE, we introduce a cohort dummy.
Di equals one if the children were born in the late birth cohort. Otherwise, it equals zero. The
regression is specified as follows:
ln yit = β0 + β1 ln yi,t−1 + β2 Di ln yi,t−1 + βX Xi + βDX Di Xi + εit
(10)
where ln y1it is the natural logarithm of child’s annual income. ln y1,t−1 is the annual lograthm
family income from both fathers and mothers over at least three years. Xi is a vector of control
variables, which include child’s age, father’s average age, quadratic forms of child’s and father’s
age, child’s gender dummy, wave dummy and provincial dummies.10
The coefficients of β1 and β2 are what we are interested in. β1 captures the IGE of income in
the early birth cohort. β2 estimates the change in IGE between early and late cohorts. β1 + β2
displays the IGE in the late cohort. Standard errors are clustered by households, as there can be
heterogeneity across different households.
9
The CHIP sample is a subsample of the national census. It follows the definition of household members from the
National Bureau of Statistics of China.
10
The reason for including father’s rather than mother’s age is because household income is mainly from fathers in
China. In addition, as father’s and mother’s ages are highly correlated, it is sufficient to include one of them only.
22
Taking the variance of income in parents’ and children’s generations into consideration, we
investigate the intergenerational income correlation (Blanden et al., 2004; Black et al., 2011). It is
defined as follows:
correlation in log income = elasticity ∗
σt−1
σt
(11)
where σt−1 and σt are the standard deviation of logarithm income of parents and children.
4.2.3
The Estimation Results
Table 5 displays the intergenerational income elasticity and correlation across early and late cohorts. The first two columns present the estimated IGE in income in each birth cohort, respectively.
Columns (3) - (4) show the corresponding intergenerational income correlation, adjusted with income variance in two generations. The last two columns present changes in the trend of IGE and
correlation. Panel 1 exhibits basic estimates by regressing children’s annual income on family
income averaged across at least three years. Panel 2 supplements the basic results by including
pre-labor market entry control variables. Gender-specific estimates are presented as well.
We find a statistically significant decrease in intergenerational income mobility from early into
late cohorts. This result is driven mainy by the female. The pattern keeps robust even including
pre-labor market entry controls into the basic regression. Specifically, the IGEs in income for all
children rises from 0.315 to 0.442 from early to late cohorts, and are both statistically significant at
the high 1% level. Ajusting the IGEs with different variance in income across the two generations,
we evidence a similar increase in the intergenerational income correlation. The rise in elasticity
and correlation is 0.127 and 0.105 respectively, with statistical significance at conventional levels
of significance.
A more striking finding lies in the gender-specific intergenerational mobility. Estimated elasticity and correlation is more than doubled for females from early into late cohorts (0.205 versus
0.496, 0.155 versus 0.373). The corresponding increase is statistically significant at the 5% level
(columns (5) - (6)). Nonetheless, we find no such statistically significant trend for males. We
23
ascribe this gender differnece to the higher increasing return to human capital for females than
males along with the market reform. We are also cautious that estimates for daughters are more
likely to be upper bound of the true value, because of coresiding restriction on the sample. To
test the robustness of the results, we include parental schooling years (Blanden et al., 2004) and
fathers’ political status (Communist Party membership) as additional controls. Estimates shrink in
magnitude, but the pattern remains robust.
To analyse the extent of possible measurement error in income, we compare results using oneyear estimates of parental income and average across at least three years in Table 6. As expected,
both of the two sets of estimates reveal consistently decreasing intergenerational mobility. However, the three-year estimates are consistently larger in magnitude. Taking the correlation for all
children for instance, compared to the one-year estimates, the three-year ones in early and late
cohorts rise by 15% and 8% respectively (columns (1) and (2)). It is because the one-year coefficients are biased by transitory fluctuations in the specific years, and thus downwarded by a factor
of σ2y /(σ2y + σ2v ),where σ2y is the variance of income in either generation, and σ2v is the variance
of transitory fluctuations around lifetime income (Solon, 1989).11 The three-year estimates yield
more convincing results.
Table 7 presents the regional disparity in intergenerational mobility. West China witnesses a
statistically significant decrease in intergenerational income mobility. East and Central China share
a similar increasing trend in IGE and correlation, though with no statistical signficance. Specifically, in the early cohort, the intergenerational income correlation in West China is 0.155 only, with
statistical significance at the 10% level. It jumps strikingly to 0.411 in the late cohort, at the high
1% level of statistical significance. The sharp increase in intergenerational correlation in economically disadvantaged western region reveals a lack of governmental subsidy on public education
and the tight budget constraint for parents to finance children’s human capital investment.12
11
The basic assumptions in Solon (1989) are that variance in income y and transitory fluctuations v are same in
either generation. In addition, v in each generation is uncorrelated with each other and with the income.
12
Detailed interpretation will be provided in the next section.
24
5
Institutional and Socioeconomic Changes and Intergenerational Mobility in China
In light of the theoretical model, this section explains our estimated time, gender, and regional
patterns of intergenerational mobility in China in the past few decades. Specifically, we discuss
how institutional, education, fiscal, and socioeconomic changes affect the theoretical determinants
of intergenerational mobility: return to human capital (r), severity of credit constraint (1 − γ), and
share of households subject to credit constraint in the theoretical framework (d). The severity of
household credit is a function of the progressivity of public investment in human capital (s) and
the price of human capital investment (p).
The decrease in the intergenerational mobility in China is mainly determined by the “fights”
among four factors. First, the return to human capital (r) have increased substantially which is
shown in Figure 6. Theoretically, the increase in return to human capital inflate the effects of intergenerational transmission of endowments and of credit constraint on intergenerational mobility.
Thus, the intergenerational mobility decreases. As discussed in the background section, the increase in return to human capital is due to the fast accumulation of physical capital, technological
progress, and, more importantly, the market-oriented institutional reforms.
Second, as shown in Figure 9, the share of governmental expenditure out of GDP has doubled
from 1992 to 2012, which increases the intergenerational mobility. Third, the educational costs
have increased dramatically since the economic reform. Figure 10 displays a sharp increase in the
share of the amount of tuition out of GDP from 1991 to 2007. The increase in educational costs
decrease the intergenerational mobility. Finally, the fast economic growth as shown in Figure 1
has led less households subject to credit constraint which increases the intergenerational mobility
in China in the past decades.
Our empirical results show the intergenerational mobility has been decreasing in China in the
past decades. Thus, the effects of the increases in both the return to human capital (r) and educational costs (p) have dominated the effects of the increases in both governmental expenditure on
25
child human capital (s) and family income (y). Therefore, the intergenerational mobility has been
decreasing in China in the past decades.
The estimation results show that the decreasing pattern of intergenerational mobility is more
evident for females. This is mainly due to two causes. First, the return to schooling years has
been higher for females than for males, and the gender gap in returning to schooling years has
been widening (Zhang, et al. 2005). Figure 11 presents the return to schooling by gender for urban
residents from 1988 to 2001. We find that the return rates to one additional schooling year are 5.2%
for females but 2.9% for males in 1988. In 2001, the return rates are 13.2% and 8.4% for females
and females, respectively. Second, girls are more subject to credit constraint in poor families due
to the residual son preference. Based on the CFPS data, Figure 12 graphs the schooling years by
gender and rural vs. urban. It shows that the gender gap in schooling years in rural areas has been
persistent. In contrast, the gender gap in schooling years in urban areas has been shrinking, and
finally it seems disappeared.
We also find a regional pattern in intergenerational mobility pattern. The decrease in intergenerational mobility in education is more evident for rural areas and western provinces. The
decrease in intergenerational income mobility is more evident for western provinces. Based on Li
et al. (2005), Figure 13 graphs the return to schooling years for six provinces. We find that the
increase in return to schooling years is more dramatically in Beijing and Zhejiang. Therefore, the
regional pattern is not driven by the return to human capital.
The regional pattern should be mainly driven by the gap in per capital income and the degree
of severity of credit constraint. On the one hand, given the same degree of the severity of credit
constraint, the intergenerational mobility in a society with higher income should be larger than
that in a society with low income. The reason is that the share of households subject to credit
constraint should be smaller. On the other hand, households in rural areas and western provinces
face a more severe credit constraint in investing their children’s human capital. As discussed in
the background section, the public finance of education is mainly localized. Therefore, the share
of government expenditure on education (s) is lower in less developed areas such as rural areas
26
and western provinces. The drastic increase in educational costs (p) has exacerbated the severity
of credit constraint more in rural areas and western provinces. Based on a recent national survey
on college students, Li (2013) shows that the poverty rates are 28% and 32% for students from
western provinces and rural areas, respectively, comparing to a national average of 22%.
Finally, we compare the pattern of intergenerational mobility in China with that in US. Figure
2 shows that the cross-sectional inequality has increased in both China and US. But the increase
in China is more dramatic. The Gini coefficients in China was only one half of that in US in 1970.
In 2008, the Gini coefficients in China and US were the same, attaining a historic high level of
0.45. The increase in cross-sectional inequality may be mainly due to the increase in the labor
productivity and then the return to human capital.
In contrast to China, Lee and Solon (2009) and Mayer and Lopoo (2004) and shows that the
intergenerational mobility in US has been increasing. The main is reason is the great increase in
federal and state government expenditure on child human capital. Since the 1970s, the US governments have initiated a series of means-tested programs targeting at alleviating the credit constraints
of child human capital investment in disadvantaged families such as Medicaid, food stamp, and
Head Start. The universal government programs also help less the severity of credit constraint.
Therefore, the positive effect of the decrease in severity in the credit constraint dominates the negative effect of the increase in return to human capital on the intergenerational mobility. In contrast,
the deterioration of the household credit constraint may reinforce the negative effect of the increase
in return to human capital on intergenerational mobility. This comparison has significant implications for designing and revising relevant pubic policies to improve the intergenerational mobility
in China, which is discussed in the section below.
6
Discussion and Conclusion
This paper studies the time, gender, and regional patterns of intergenerational mobility in education
and income in China. Our empirical results show that intergenerational mobility in both education
27
and income has been decreasing. Specifically, one year of parental schooling leads to additional
0.37 schooling years for children born in 1975-1979, which is 0.33 years for children born in
1965-1969. The intergenerational income elasticity increases from 0.32 for cohorts born before
1970 to 0.44 for cohorts born after 1970. The decreasing pattern is more evident for females, rural
residents, and residents from western provinces. We further explore mechanisms translating the
market-oriented institutional reforms and socioeconomic changes to the estimated patterns under
a unified economic theory. The decrease in intergenerational mobility is driven by the increases
in both return to human capital and educational costs. The gender disparity is due to the gender
difference in return to human capital. The regional disparity is caused by the decentralization in
government financing education.
Based on our estimates of the intergenerational mobility, the cross-sectional inequality is deteriorated for the younger generation, especially for females, rural residents, and residents from
western regions in China. We know that
ln yt = µ + β ln yt−1 + v.
When it is at the steady state that var(yt ) = var(yt−1) ,
var(ln yt ) =
var(v)
.
1 − β2
The smaller the β, the lower the intergenerational mobility, and then the higher the cross-sectional
inequality at the steady state. Therefore, any factor decreasing the intergenerational mobility will
increase cross-sectional inequality in the long run. This result holds in the general case when the
economy is not at the steady state.
The intergenerational mobility in China in the future mainly depends on the “fights” among
the return to education and the severity of the household budget constraint in investing the child’s
education. The return to education depends on the interaction between the demand and supply
28
of higher education. With the accumulation of the physical capital, the technological progress,
and further market-oriented economic reform, the demand for high-skilled labor increases in the
future. On the other hand, the supply of college graduates increases substantially with the higher
education. We conjecture that the supply shock of the fresh college graduates to the labor market
would be temporarily. In the long run, the demand for high-skilled labor dominates. There will be
a steady increase in the return to education, which decreases in the intergenerational mobility in
the future.
To promote intergenerational mobility in China, the Chinese government should make prompt
and effective actions to lessen the household credit constraint. As discussed in the previous comparison between China and US, we should initiate various programs to subsidize the education of
those children from disadvantaged families, such as those left-behind children. During the same
time, the efficacy of the loan and scholarship programs at the tertiary level should be enhanced.
More importantly, the central government should allocate a larger share of its budget to invest the
child’s human capital and to enhance the efficiency of the usage of this budget. China has made
substantial progress in these regards, but more efforts are needed to ensure equal access to qualified
education for all children.
Finally, although intergenerational mobility in China is very importantly in both academic and
policy analysis, there are few rigorous analysis in this regard. Future research on this topic is highly
desirable. Besides the intergenerational mobility through monetary human capital investment, the
intergenerational transmission of wealth and political status are in our research agenda in the future.
29
References
30
.25
.3
.35
Gini coefficient
.4
.45
Real per capita GDP (2005 I$/person)
8000
2000
4000
6000
0
1980
1990
2000
2010
Year
Real per capita GDP (2005 I$/person)
Gini coefficient
Data source: Penn World Table (version 7.1) and NBS (2012)
Figure 1: The Per Capita GDP and Gini Coefficient in China
.7
Gini coefficients
.6
.4
.5
.3
1970
1980
1990
Brazil
USA
Japan
2000
China
Germany
France
Figure 2: Gini Coefficients: International Comparison
32
2008
3.5
Urban/rural (per capital income)
2
2.5
3
1.5
1980
1990
2000
Year
Data source: NBS (2012)
Figure 3: Real Per Capital Income: Rural vs. Urban
33
2010
12
10
8
6
4
2
1988
1990
1992
1994
Year
Beijing
Zhejiang
Shaanxi
1996
1998
Liaoning
Guangdong
Sichuan
Data source: Li et al. (2005)
Figure 4: Regional Disparity in Real Per Capital Income
34
2000
6000
5000
4000
3000
2000
1000
1990
1995
2000
Year
Low-education level
High-education level
2005
Medium-education level
Data source: Li et al. (2012)
Figure 5: Annual Wage of Urban Workers
35
2010
50
40
30
20
10
0
1990
1995
2000
Year
Returns to schooling years
2005
Returns to college education
Figure 6: Return to Education in Urban China
36
2010
100
80
60
40
20
1980
1990
2000
2010
Year
Senior secondary
Primary
Junior secondary
Data source: Knight, Sicular, and Yue (2012)
Figure 7: The increase in primary school enrollment rates and secondary school progression rates
37
22
20
18
16
14
1988
1990
1992
1994
1996
Year
Females
1998
Males
Data source: Li and Xing (2010)
Figure 8: Higher Education Expansion
38
2000
2002
Government Educational Expenditure / GDP
.035
.02
.025
.04
.03
1992
1994
1996
1998
2000
2002
Year
2004
2006
2008
Data source: NBS (2013)
Figure 9: Governmental Expenditure on Education/GDP
39
2010
2012
.008
.4
0
.002
.004
.006
Tuition/GDP
Tuition/Gov. Edu. Exp
.1
.2
.3
0
1991
1993
1995
1997
1999
Year
Tuition/Gov. Edu. Exp
2001
2003
2005
Tuition/GDP
Data source: NBS (2013)
Figure 10: The increase in tuition in China
40
2007
15
10
5
0
1988
1990
1992
1994
Year
Women
1996
1998
Men
Data source: Zhang et al. (2005)
Figure 11: Return to schooling years by gender
41
2000
12
10
8
6
4
1965
1970
1975
Birth year
Rural males
Urban males
Rural females
Urban females
Data source: CFPS 2010
Figure 12: Schooling years by gender and by region
42
1980
12
10
8
6
4
2
1988
1990
1992
1994
Year
Beijing
Zhejiang
Shaanxi
1996
1998
Liaoning
Guangdong
Sichuan
Figure 13: Return to schooling years by province
43
2000
Table 1: Summary Statistics of the CFPS data
Overall
Cohort 1: 1965-1969
childs schooling years
parental schooling years a
childs gender (male=1)
childs Hukou status
(agricultural=1)
annual family income b
Observation
Cohort 2: 1970-1974
childs schooling years
parental schooling years a
childs gender (male=1)
childs Hukou status
(agricultural=1)
annual family income b
Observation
Cohort 2: 1975-1979
childs schooling years
parental schooling years a
childs gender (male=1)
childs Hukou status
(agricultural=1)
annual family income b
Observation
a
b
Mean (Standard deviation)
Urban
Rural
East
Central
West
7.144
(4.399)
2.794
(3.654)
0.497
(0.500)
0.706
(0.456)
5381.5
(15484.0)
3,734
10.612
(3.612)
4.641
(4.406)
0.508
(0.500)
9124.5
(14744.7)
1,097
5.702
(3.859)
2.025
(2.971)
0.492
(0.500)
3521.1
(15511.3)
2,637
7.543
(4.201)
3.063
(3.844)
0.487
(0.500)
0.694
(0.461)
5736.8
(12208.1)
1,201
7.637
(4.306)
3.060
(3.747)
0.493
(0.500)
0.642
(0.479)
5803.5
(19446.2)
1,502
5.962
(4.539)
2.093
(3.167)
0.514
(0.500)
0.702
(0.458)
4378.0
(11579.4)
1,031
7.556
(4.463)
3.382
(3.692)
0.497
(0.500)
0.694
(0.461)
6204.0
(12777.9)
4,393
11.140
(3.664)
5.643
(4.219)
0.493
(0.500)
10090.9
(14976.2)
1,343
5.977
(3.825)
2.387
(2.925)
0.498
(0.500)
4334.1
(11102.3)
3,050
8.274
(4.215)
3.895
(3.870)
0.481
(0.500)
0.666
(0.472)
6927.4
(13143.6)
1,393
7.941
(4.219)
3.681
(3.696)
0.494
(0.500)
0.633
(0.482)
6071.6
(13420.6)
1,738
6.232
(4.767)
2.406
(3.279)
0.517
(0.500)
0.810
(0.393)
5653.7
(11453.9)
1,262
8.231
(4.423)
3.822
(3.756)
0.486
(0.500)
0.703
(0.457)
7552.5
(13737.0)
3,681
11.798
(3.484)
6.403
(3.995)
0.480
(0.500)
11643.5
(17771.3)
1,094
6.723
(3.879)
2.731
(3.053)
0.490
(0.500)
5903.3
(11319.7)
2,587
9.212
(3.911)
4.168
(3.785)
0.490
(0.500)
0.678
(0.467)
10010.0
(16976.3)
1,186
8.830
(4.098)
4.434
(3.857)
0.486
(0.500)
0.626
(0.484)
6962.4
(13171.9)
1,444
6.302
(4.787)
2.592
(3.265)
0.482
(0.500)
0.836
(0.370)
5536.8
(9158.9)
1,051
: Parental schooling years refer to the average schooling years between fathers and mothers.
: Annual family income refers to the total annual income from fathers and mothers.
44
45
0.327***
(0.0182)
0.300***
(0.0235)
0.360***
(0.0253)
0.360***
(0.0166)
0.304***
(0.0229)
0.413***
(0.0230)
0.374***
(0.0182)
0.327***
(0.0254)
0.417***
(0.0232)
1970-1974 1975-1979
(2)
(3)
0.272***
(0.0151)
0.26***
(0.0204)
0.293***
(0.0206)
1965-1969
(4)
0.298***
(0.0137)
0.261***
(0.0197)
0.334***
(0.0186)
1970-1974
(5)
Correlationa
0.318***
(0.0155)
0.290***
(0.0225)
0.344***
(0.0191)
1975-1979
(6)
0.033
(0.0246)
0.004
(0.0328)
0.053
(0.0342)
(2) - (1)
(7)
0.047**
(0.0257)
0.027
(0.0346)
0.057**
(0.0343)
(3) - (1)
(8)
Change in coefficient
0.026
(0.0204)
0.001
(0.0283)
0.041
(0.0277)
(5) - (4)
(9)
0.046**
(0.0216)
0.030
(0.0304)
0.051**
(0.0281)
(6) - (4)
(10)
Change in correlation
Note: The dependent variable is childs schooling years. The independent variable is parental average schooling years. The control
variables include age of children, average age of parents, Hukou status (agricultural or non-agricultural), and gender dummy in the
specification for all children. Data source: Chinese Family Panel Studies 2010. Standard errors clustered by households are in brackets;
* significant at 10%; ** significant at 5%; *** significant at 1%.
a
: Intergenerational educational correlation = intergenerational educational coefficient ∗σ p /σc , where σ p and σc are the standard
deviation of schooling years of parents and children, respectively.
Daughters
Sons
All children
1965-1969
(1)
Regression coefficient (β)
Table 2: Intergenerational Educational Mobility in China by Birth Cohorts
46
0.288***
(0.0248)
0.368***
(0.0261)
0.325***
(0.0302)
0.296***
(0.0274)
0.369***
(0.0414)
0.312***
(0.0234)
0.404***
(0.0232)
0.301***
(0.0277)
0.336***
(0.0252)
0.421***
(0.0365)
0.299***
(0.0268)
0.428***
(0.0239)
0.294***
(0.0259)
0.324***
(0.0280)
0.496***
(0.0415)
0.351***
(0.0303)
0.283***
(0.0201)
0.297***
(0.0276)
0.258***
(0.0238)
0.257***
(0.0289)
0.359***
(0.0269)
0.309***
(0.0177)
0.276***
(0.0254)
0.294***
(0.0221)
0.290***
(0.0251)
0.343***
(0.0307)
0.337***
(0.0188)
0.285***
(0.0251)
0.305***
(0.0264)
0.338***
(0.0283)
0.024
(0.0341)
0.036
(0.0349)
-0.024
(0.0410)
0.04
(0.0370)
0.052
(0.0552)
0.011
(0.0365)
0.06**
(0.0354)
-0.031
(0.0398)
0.028
(0.0391)
0.127**
(0.0586)
Correlationa
Change in coefficient
1965-1969 1970-1974 1975-1979 (2) - (1)
(3) - (1)
(4)
(5)
(6)
(7)
(8)
0.008
(0.0405)
0.026
(0.0268)
-0.021
(0.0376)
0.036
(0.0325)
0.033
(0.0383)
-0.008
(0.0431)
0.054**
(0.0275)
-0.012
(0.0373)
0.047
(0.0355)
0.081**
(0.0404)
Change in correlation
(5) - (4)
(6) - (4)
(9)
(10)
Note: The dependent variable is childs schooling years. The independent variable is parental average schooling years. The control
variables include age of children, average age of parents, Hukou status (agricultural or non-agricultural), and gender dummy in the
specification for all children. Data source: Chinese Family Panel Studies 2010. Standard errors clustered by households are in brackets;
* significant at 10%; ** significant at 5%; *** significant at 1%.
a
: Intergenerational educational correlation = intergenerational educational coefficient ∗σ p /σc , where σ p and σc are the standard
deviation of schooling years of parents and children, respectively.
West
Central
East
Rural
Urban
Regression coefficient (β)
1965-1969 1970-1974 1975-1979
(1)
(2)
(3)
Table 3: Intergenerational Educational Mobility in China by Regions
Table 4: Summary Statistics of the CFPS data
Mean
(Standard deviation)
East
Central
Overall
Early cohort
childs annual income
annual family income a
childs gender (male=1)
childs age
fathers age
Observation
Late cohort
childs annual income
annual family income a
childs gender (male=1)
childs age
fathers age
Observation
West
6628.004
(5783.354)
9331.364
(5887.859)
0.705
(0.456)
29.523
(4.298)
57.139
(4.683)
627
8572.219
(6822.393)
10899.52
(6497.635)
0.679
(0.468)
30.072
(4.473)
57.467
(4.609)
293
4827.629
(3040.306)
8010.927
(5380.898)
0.722
(0.449)
29.269
(4.103)
57.151
(4.693)
212
5087.22
(5231.885)
7859.751
(3972.818)
0.738
(0.442)
28.648
(4.041)
55.334
(4.782)
122
8939.762
(7622.202)
15432.46
(11384.180)
0.6
(0.490)
25.462
(2.353)
53.28
(4.548)
821
11502.71
(8973.904)
18809.11
(14051.720)
0.608
(0.489)
25.632
(2.407)
53.378
(4.163)
383
6567.196
6888.833
(4816.738) (5885.124)
11958.42
13234.25
(7029.896) (7396.300)
0.61
0.57
(0.489)
(0.496)
25.154
25.542
(2.183)
(2.441)
53.098
53.334
(4.917)
(4.793)
259
179
Note: Early cohort includes children born between 1949 (the foundation of People’s Republic of
China) and 1970 (included). Late cohort includes children were born after 1970. They received
education and worked in the post 1978 economic reform era. Income is converted to the 2002 Yuan
using Consumer Price Index.
a
: Annual family income refers to the average annual income from fathers and mothers in the
proceeding three years (at least) before the survey wave (included).
47
48
0.336***
(0.0336)
0.319***
(0.0621)
0.373***
(0.0554)
0.323***
(0.0379)
0.314***
(0.0482)
0.362***
(0.0634)
0.231***
(0.0461)
0.241***
(0.0476)
0.155*
(0.0880)
0.216***
(0.0509)
0.236***
(0.0528)
0.101
(0.0952)
Correlationa
Early cohortb Late cohortb
(3)
(4)
0.129
(0.0848)
0.0804
(0.0957)
0.347**
(0.1520)
0.127*
(0.0754)
0.0812
(0.0861)
0.291**
(0.1370)
Change in
coefficient
(5)
0.107*
(0.0634)
0.078
(0.0715)
0.261**
(0.1140)
0.105**
(0.0571)
0.078
(0.0783)
0.218**
(0.1040)
Change in
correlation
(6)
Note: Children are at least 23 years old. Fathers are less than 65 years old. Income is converted to the 2002 Yuan using Consumer Price
Index. Data source: Chinese Household and Income Projects 1995 and 2002 in urban China. Standard errors clustered by households are
in brackets; * significant at 10%; ** significant at 5%; *** significant at 1%.
a
: Intergenerational income correlation = intergenerational income coefficient ∗σ p /σc , where σ p and σc are the standard deviation of
logarithm annual income of parents and children, respectively.
b
: Early cohort includes children born between 1949 (the foundation of Peoples Republic of China) and 1970 (included). Late cohort
includes children born after 1970. They received education and worked in the post 1978 economic reform era.
c
: The dependent variable is the logarithm of childs annual income. The independent variable is the average of the logarithm of annual
family income over at least preceding three years. The control variables include age and age squared of children and fathers, gender
dummy in the specification for all children, wave dummies and provincial dummies.
d
: In augmented regressions, additional control variables include fathers Communist Party membership, and average schooling years
between fathers and mothers.
Regression coefficient (β)
Early cohortb Late cohortb
(1)
(2)
1. Income regressionsc
All children
0.315***
0.442***
(0.0630)
(0.0443)
Sons
0.335***
0.416***
(0.0661)
(0.0811)
Daughters
0.205*
0.496***
(0.1160)
(0.0738)
2. Augmented income regressionsd
All children
0.297***
0.426***
(0.0700)
(0.0499)
Sons
0.330***
0.410***
(0.0737)
(0.0631)
Daughters
0.135
0.482***
(0.1270)
(0.0844)
Table 5: Intergenerational Income Mobility in China by Birth Cohorts
49
0.311***
(0.0287)
0.296***
(0.0391)
0.339***
(0.0436)
0.336***
(0.0336)
0.319***
(0.0621)
0.373***
(0.0554)
0.201***
(0.0462)
0.221***
(0.0465)
0.115
(0.0876)
0.231***
(0.0461)
0.241***
(0.0476)
0.155*
(0.0880)
Correlationa
Early cohortb Late cohortb
(3)
(4)
0.127*
(0.0754)
0.0812
(0.0861)
0.291**
(0.1370)
0.164**
(0.0737)
0.0918
(0.0842)
0.353***
(0.1250)
Change in
coefficient
(5)
0.105**
(0.0571)
0.078
(0.0783)
0.218**
(0.1040)
0.110**
(0.0544)
0.0751
(0.0607)
0.224**
(0.0978)
Change in
correlation
(6)
Note: Children are at least 23 years old. Fathers are less than 65 years old. Income is converted to the 2002 Yuan using Consumer Price
Index. Data source: Chinese Household and Income Projects 1995 and 2002 in urban China. Standard errors clustered by households are
in brackets; * significant at 10%; ** significant at 5%; *** significant at 1%.
a
: Intergenerational income correlation = intergenerational income coefficient ∗σ p /σc , where σ p and σc are the standard deviation of
logarithm annual income of parents and children, respectively.
b
: Early cohort includes children born between 1949 (the foundation of Peoples Republic of China) and 1970 (included). Late cohort
includes children born after 1970. They received education and worked in the post 1978 economic reform era.
c
: The dependent variable is the logarithm of childs annual income. The independent variable is the average of the logarithm of annual
family income over at least preceding three years. The control variables include age and age squared of children and fathers, gender
dummy in the specification for all children, wave dummies and provincial dummies.
d
: In augmented regressions, additional control variables include fathers Communist Party membership, and average schooling years
between fathers and mothers.
Regression coefficient (β)
Early cohortb Late cohortb
(1)
(2)
1.One-year estimatesc
All children
0.275***
0.439***
(0.0633)
(0.0406)
Sons
0.317***
0.409***
(0.0668)
(0.0541)
Daughters
0.142
0.495***
(0.1080)
(0.0637)
d
2.Three-year estimates
All children
0.315***
0.442***
(0.0630)
(0.0443)
Sons
0.335***
0.416***
(0.0661)
(0.0811)
Daughters
0.205*
0.496***
(0.1160)
(0.0738)
Table 6: Intergenerational Income Mobility in China: Sensitivity Analysis
50
0.266***
(0.0559)
0.194**
(0.0972)
0.155*
(0.0864)
0.341***
(0.0488)
0.293***
(0.0600)
0.411***
(0.0740)
Correlationa
Early cohortb Late cohortb
(3)
(4)
0.099
(0.0937)
0.130
(0.1520)
0.321**
(0.1550)
Change in
coefficient
(5)
0.075
(0.0743)
0.0995
(0.1140)
0.256**
(0.1140)
Change in
correlation
(6)
Note: Children are at least 23 years old. Fathers are less than 65 years old. Income is converted to the 2002 Yuan using Consumer Price
Index. Data source: Chinese Household and Income Projects 1995 and 2002 in urban China. Standard errors clustered by households are
in brackets; * significant at 10%; ** significant at 5%; *** significant at 1%.
a
: Intergenerational income correlation = intergenerational income coefficient ∗σ p /σc , where σ p and σc are the standard deviation of
logarithm annual income of parents and children, respectively.
b
: Early cohort includes children born between 1949 (the foundation of Peoples Republic of China) and 1970 (included). Late cohort
includes children born after 1970. They received education and worked in the post 1978 economic reform era.
c
: The dependent variable is the logarithm of childs annual income. The independent variable is the average of the logarithm of annual
family income over at least preceding three years. The control variables include age and age squared of children and fathers, gender
dummy in the specification for all children, wave dummies and provincial dummies.
d
: In augmented regressions, additional control variables include fathers Communist Party membership, and average schooling years
between fathers and mothers.
Regression coefficient (β)
Early cohortb Late cohortb
(1)
(2)
a
Three-year estimates
East
0.339***
0.438***
(0.0713)
(0.0628)
Central
0.261**
0.391***
(0.1310)
(0.0800)
West
0.224*
0.545***
(0.1250)
(0.0982)
Table 7: Intergenerational Income Mobility in China by Regions