Enforcing Government Policy: Privatization and the
Weakening Effects of China’s One-child Policy∗
Hua Cheng†
December 28, 2015
Abstract
I evaluate the role of China’s privatization of the economy in weakening the effects of
the birth control policy, known as the one-child policy in the West, during the last three
decades. State-owned enterprises or other state-owned organizations (SOEs) imposes more
control over individual choices and hence represents increased state capacity (Acemoglu
et al., 2015). The baseline OLS regressions show that compared to those not working in a
SOE, the number of children for individuals working in a SOE is 0.1 lower on average. In
other words, if the individuals not working in SOEs came to SOEs, we would probably lose
10 children for each one hundred of them. To deal with the selection issue, I use father’s
SOE status as an instrumental variable. The 2SLS results show a much larger difference in
the number of children: 0.4. The existence of positive selection and the reduction in bias
from job type switches probably rationalize the larger 2SLS results. Furthermore, Probit
regressions give a strong negative impact of working in SOEs on the probability to have
more than one child. These results imply that the birth control policy might already lead to
an even lower fertility rate in China without privatization.
∗I
am indebted to Daniel S. Hamermesh, Kishore Gawande and Michael Geruso for their guidance and support.
I am also grateful to Yan Dong, Shihe Fu, Sukjin Han, Robert Jensen, Han Li, Leigh Linden, Xiaobo Lu, Xin
Meng, Richard Murphy, Stephen J. Trejo, Xiao Wei, Emily Weisburst, Haiqing Xu, Yang Yao and participants in
2016 AEA annual meeting, 2015 China Economics Annual Conference, Southwestern University of Finance and
Economics and UT-Austin Public/labor seminar for helpful comments and discussions. The main data used in this
paper were collected by the research project “Chinese General Social Survey (CGSS)” carried out by the National
Survey Research Center, Renmin University of China (NSRC). The author appreciate the assistance in providing
data by the institutes and individuals aforementioned. The views expressed herein and the possible errors are the
author’s own.
† Department of Economics, University of Texas at Austin. E-mail address: [email protected]
1
1
Introduction
China’s birth control policy, known as the one-child policy in the West, is perhaps a policy
affecting the largest number of births in history. China’s national fertility dropped below the
replacement level in early 1990s and continues its downward trend ever since, with a total fertility rate (TFR) at about 1.7 births per woman in 2012 according to the World Bank1 . The 2000
census gave a lower TFR of 1.22 and the 2010 census reported a TFR of 1.18, which are only
one half of the world average. Although there is a disagreement over how much each factor
contributes, a consensus is that both the birth control policy and socioeconomic changes lead
to such a sharp decline in fertility. The importance of China’s fertility comes not only from its
sheer population size, one-fifth of world population, but also from the integration of this country
into the global economic system. As the “world factory” and the largest exporter in the world,
China provides cheap products to global consumers and keeps a low inflation around the world,
so that China’s demographic change will affect the global economy profoundly. Furthermore,
the declining fertility rate combined with population aging implies a sharply increasing dependency ratio of population in near future, which in turn will threaten the social welfare system in
this country.
In this paper, I argue that there is another economic force that already weakens the effects of
the birth control policy and slows down the fertility decline, which has not yet been discussed
by the previous literature. State-owned enterprises and other state-owned organizations (hereafter referred to as the SOEs), imply more control over individual choices by the government.
Motivated by this reasoning, I investigate if an individual working in a SOE will have less children than his counterparts working in a non-SOE. Calculated based on China Labour Statistical
Yearbooks, as of 1978, there were about 78% of urban labor working in SOEs2 . The share
of urban labor in SOEs declined to 18% in 20103 . Although there was a large change in the
composition of urban labor because of the inflow of rural migrant workers, the share of urban
labor in SOEs still decreased sharply excluding rural migrant workers.
Using a novel repeated cross-sectional data set, I find a strong and economically meaningful
effect of working in a SOE on the fertility rate, in linear regressions, 2SLS regressions using
1 See
World Bank database at http://data.worldbank.org/indicator/SP.DYN.TFRT.IN.
remaining 22% were unemployed or in collective enterprises, nominally owned by inside workers, while
there were very few individuals in private firms or foreign-owned firms. The data can be find on the website of
National Bureau of Statistics of China: http://data.stats.gov.cn/workspace/index?m=hgnd. As a caution, this share
of urban labor working in SOEs is based on a smaller size of urban areas and urban population because of the
urbanization in the last three decades.
3 Although after 2005 there is a revival of SOEs perhaps due to a large-scale infrastructure construction and the
jump in the prices of natural resources, where big SOEs dominate, the share of SOEs in urban employment still
continued to drop from 22% in 2005 to 18% in 2010.
2 The
2
father’s SOE status as an instrumental variable, Probit and IV Probit regressions. This implies
that the previous research probably underevaluate the potential impacts of the birth control
policy. Without privatization of the economy during the last three decades, we would probably
see an even lower fertility rate in China. Since SOEs already take up a much smaller share of
the total employment than three decades ago and their control over individual choices are also
weakened than before, the potential of privatization in slowing down the decline in fertility is
limited. These facts suggest a strong policy implication: given a low fertility rate in China, a
further relaxation or abolition of the birth control policy is needed.
The goal of this paper is to make progress on understanding the role of the government and
economic system in affecting the fertility rate by providing evidence for the hypothesis that the
government control over economy can directly affect citizens’ fertility decisions. The SOEs
can be seen as increased state capacity (Acemoglu et al., 2015), the ability of the government to
enforce its policies, over their employees. Acemoglu et al. (2015) document positive impacts
of such state capacity on Columbia’s local economic growth and educational attainment. In my
paper, I focus on the impacts of state capacity on family decisions. Although order is perhaps
more important in Columbia because of long-term military conflicts, the strong state capacity in
China is not necessarily always productive, and hence the weakening control of the government
over citizens should be more likely to have positive effects.
In the major parts, I keep the male sample with urban residence certificate (“Hukou”) age
18-60, married or once married. The males are usually the primary earner in a family and hence
the job type should have a larger impact on them, though most of the results for the female sample are qualitatively identical. In China, there are differential family planning policies toward
rural residents and urban residents based on residence certificate. In general, rural residents are
allowed to have a second child if the first one is a girl, which is partly attributed to the agricultural production need for male labor. However, the official one-child restriction is quite rigid
and uniform among the citizens with urban residence certificate, especially those in the same
province. That is, although plenty of rural migrant workers work in urban areas, they are not
restricted by the stricter control over fertility since they usually do not have urban residence
certificate. The baseline OLS regressions show that compared to those not working in a SOE,
the number of children for individuals working in a SOE is about 0.1 lower on average. In
other words, if the individuals not working in SOEs came to SOEs, we would probably lose 10
children for each one hundred of them.
Several thorny issues might prevent effective identification if left unsolved. First, I can
obtain the number of all children an individual has rather than the number of children living
with the individual, as that contained in many other datasets. Secondly, a worker might switch
3
from a SOE job to a non-SOE job or do the converse. Resulting from the SOE and non-SOE
segmentation in China (Knight and Yueh, 2009; Wang et al., 2014), the number of job type
switches is not large; especially, in Section 3.2 I explain why these job type switches lead to
the underestimation bias and how my IV strategy using father’s SOE status as an instrumental
variable reduces this bias. Thirdly, SOE workers and non-SOE workers are not the same, that
is, there is a selection issue. For example, it is possible that people with a weaker preference
toward children or a higher ability to raise children are more likely to work in SOEs, so that
the selection might be positive or negative. I find that positive selection seems more important,
and my IV strategy also solves this selection issue. I provide extensive historical evidence and
empirical evidence from regressions to support the use of this instrumental variable.
Besides the usual reasons for using father’s characteristics, my practice is based on a special institutional feature in China, which leads to a strong impact of father’s job type on their
children’s. For a long period till the late 1990s, children were usually allowed to take their
parents’ jobs in SOEs one-by-one (Dingzhi). For example, if both of the parents are working
in SOEs and they have two children, then both of the children can enter the same firms upon
their parents’ retirement or early retirement. In addition, the exclusion constraint is plausibly
satisfied because in general the group of fathers spent most of their career during the planned
economy era and had little freedom to choose their jobs, and there is not evidence showing that
the government distributed SOE jobs to groups with certain preference toward children. The
2SLS results show that compared to those not working in a SOE, the number of children for
individuals working in a SOE is 0.4 lower on average. Finally, in regressions with a dummy
variable for one child or more than one child as the dependent variable, I also find a strong
negative impact of working in SOEs on the probability to have more than one child.
The structure of the paper is as follows. Section 2 provides an introduction to the historical
background of the birth control policy and a review of the related literature. Section 3 presents
the empirical strategy; Section 4 describes the dataset; Section 5 and Section 6 demonstrates
the results and robustness check. Finally, Section 7 concludes.
2
2.1
Background and Literature Review
Background
Although the declining fertility trend in China is broadly consistent with the experiences of developed countries and newly industrialized countries accompanied by economic development, it
is also widely considered as a special case related to government intervention. Beginning in the
4
early 1970s, the government initiated its “Wan, Xi, Shao”(later, longer, fewer) policy. And in
1979, grounded in a neo-Malthusian concern over the negative effects of a large population size
on economic growth, the government launched the much radical and controversial one-child
policy. Supported by a well-established bureaucracy devoted to routine surveillance and policy
enforcement, the policy penetrates Chinese society from the highest level of the government
down to urban neighborhoods and rural villages. Even with its later adjustments and modifications, the draconian “one-child per couple” rule applies to nearly two-thirds of Chinese couples
till recently (Gu et al., 2007).
The birth control policy is nationwide, however, provinces have some autonomy in implementing this policy (Cai, 2010). Beginning in 1987, official policy granted local officials the
flexibility to make exceptions and allow a second children in the case of “practical difficulties”
such as cases in which the father is a disabled serviceman, and some provinces had other exemptions worked into their policies as well. Families with children with disability face different
policies and families whose first child suffers from physical disability, mental illness, or intellectual disability are allowed to have more children. The number of people in these groups is
very small and should not drive a statistically significant result in my sample.
Starting from 1984, in most rural areas, families are allowed to apply to have a second child
if their first-born is a daughter, which is partly attributed to the importance of male labor in
agricultural production. Notice that the differential policies toward rural residents and urban
residents are based on residence certificate. That is, although plenty of rural migrant workers
work in urban areas, they are not restricted by the stricter control over fertility in urban areas
since they usually do not have urban residence certificate. Because of this institutional difference and the fact there exist very few SOEs in rural areas4 , I will not include the individuals
with rural residence certificate.
Furthermore, starting from the late 1990s, the majority of provinces and cities permit two
parents who were “only children” themselves to have two children. In 2013, this rule was
relaxed further: couples in which one parent is an only child are allowed to have a second
child5 . These policies are relatively new and thus have no impact on most individuals in my
sample. Moreover, it is rare to see both couples are the only child themselves. In addition, since
people working in SOEs are more likely to be the only child (as found in my paper) and can
enjoy the relatively relaxed policies, my estimates are probably conservative. Finally, ethnic
minority groups face relatively relaxed restrictions. I will control the dummy variable for ethnic
4 SOEs
are usually located in urban areas. Furthermore, at least before and in 1980s, when the government
brought rural labor into SOEs, it usually grant urban residence certificate together.
5 For more information about this new policy, see “The Economist Explains - Why is China relaxing its onechild policy?” on the Economist, January 27, 2015.
5
Han or not to account for this impact. A more comprehensive introduction to the birth control
policy can be seen in Fong (2006).
Usually, a job in SOEs is more attractive than most other jobs. It usually means a high level
of job security, better retirement benefits even if the regular wage level is not always higher than
that in many non-SOEs, such as foreign invested firms, especially before the large-scale SOE
reform in the late 1990s. And importantly, a SOE is much more likely to grant the local urban
residence certificate, which is related to many kinds of valuable social welfare, such as better
health care access and better education for their children. These facts mean that many people
probably tolerate the stricter restrictions on their fertility decisions. The benefits from working
in a SOE increases the opportunity cost of an extra child, and hence lead to a smaller number
of children in the framework of Becker (1960). Although there is not a nationwide official
law stating the stronger restrictions in SOEs, in order to implement the one-child policy, a
SOE can effectively threat its workers in terms of reduced wage, fringe benefits and promotion
opportunities, and even fire them. The promotion of SOE managers is directly related to the
effectiveness of the one-child policy in their territory, hence it is not surprising that many of
them have a strong incentive to enforce this policy.
Extensive media coverage provides stories about the impacts of working in a SOE on fertility decision. A recent report by the Beijing Youth Daily quotes a local official in Hunan
province in National Health and Family Planning Commission (NHFPC), “Most urban families
desiring two children already evaded the birth control through the one and a half child policy
or just gave birth to more children. The main affected group by the one-child policy is those
working in state-owned enterprises or other state-owned organization” 6 . Another news covered
by Chongqing Evening News covered a story that a public school teacher was fired and forced
to pay a penalty, called the social maintenance fee, of 50 thousand yuan, more than threefold
of the per capita disposable income in urban Chongqing7 . In contrast to these, it is rarely seen
a worker in a private firm is fired because of an additional birth, not to mention those selfemployed. However, to the best of my knowledge, there is not a study directly approaching the
role of privatization in weakening the birth control policy, probably resulting from the thorny
issues mentioned in the introduction. My paper will fill this blank.
6 NHFPC:
A Unrestricted “Two children policy” has not been put on agenda, May 7, 2015, see
http://news.ynet.com/3.1/1505/07/10052917.html.
7 A public school teacher was fired and forced to pay a social maintenance fee of 50 thousand yuan, March 29,
2008, see http://news.sohu.com/20080329/n255977332.shtml.
6
2.2
Related Literature
The SOEs can be seen as increased state capacity (Acemoglu et al., 2015), the ability of the
government to enforce its policies, over their employees, and hence my paper extends Acemoglu
et al. (2015)’s discussion into the impacts of state capacity on family decisions. State capacity
is linked to the “East Asian Miracle”(Johnson, 1982; Amsden, 1989; Wade, 1990; Evans, 1995)
and the economic failure of African or Latin American nations (Herbst, 2000; Centeno, 2002).
Cross-country empirical evidence presented in Gennaioli and Rainer (2007) and within-country
evidence in Michalopoulos and Papaioannou (2013) and Bandyopadhyay and Green (2012) also
show a positive association between measures of historical political centralization and presentday outcomes. More discussions on state capacity can be found in Acemoglu et al. (2015).
Because state ownership is common in countries other than the U.S. (i.e. La Porta et al., 2002),
including European countries and many other emerging economies, my results can generalize.
Broadly, my paper is related to three strands of literature. The first is the literature about
the impacts of privatization in transitional economies. Starting from the 1980s, Soviet Union,
Eastern European countries, China, India and many other countries experience a transition from
a centrally planned economy to a market economy, with a push to privatize state-owned enterprises and resources, and a blossom of newly established private enterprises. A huge literature
documents that privatization improves corporate performance (Frydman et al., 1999), increases
bank efficiency (Bonin et al., 2005) and wage level (Brainerd, 2002), discusses the relationship between methods of privatization and economic growth (Bennett et al., 2007). LastarriaCornhiel (1997) explores the transformation of customary tenure systems and their impact on
women’s rights to land in Africa. Estrin et al. (2009) summarize the studies on the effects of
privatization and ownership in transition economies.
However, in former communist countries, to a less extent in many developing countries,
SOEs are not only used by the government to control economic resources, but also are an effective instrument to impose restrictions on individual choices. During the New Economic
Policy period (early 1920s) in Russia and the early communist period in China (early 1950s), a
vivid private economy existed but was soon taken over by the governments. If the government
just taxed the private enterprises, it might already achieve a high economic growth and obtain
enough tax revenue like what happens in the United States, rather than tolerated a long-term
stagflation. This implies that economic performance is probably not the main concern for these
governments. In fact, in Mao’s China (1950s to 1970s), a couple needed to get approved by the
organizations they worked before marrying. Till now, although the share of SOEs in the national
production and employment is much lower than three decades ago, they still dominate in most
profitable industries and usually pay a higher wage and fringe benefits to their employees com7
pared to most private firms, besides more job security. This compensation means that at least a
share of individuals would like to tolerate more restrictions on their individual choices including
fertility decisions, a rational choice based on the cost-benefit analysis. My paper deepens the
understanding on the effects of the government control through SOEs and its relaxation.
This study also adds to the literature about the determination of fertility in China. The
Chinese government8 , quoting Zhenwu Zhai, director of Renmin University’s School of Sociology and Population in Beijing, estimates that 400 million births were prevented by the onechild policy as of 2011. This estimate probably underevaluates the impacts of socioeconomic
changes. Several studies have demonstrated that fertility variation in China at the provincial
level is closely related to variations in economic and social development (e.g., Poston and Gu
1987; Poston 2000; Chen et al. 2009; Cai, 2010). A long-term experiment in a county in Shanxi
Province where the family planning law was suspended, suggests that families would not have
many more children even if the law were abolished. The problem is that this experiment is limited to just a county, and hence it is not clear the nationwide impacts would be. However, while
almost all these research attempt to find the reasons of a fertility decline, my paper provide a
mechanism in slowing down the decline. That is, many previous studies may underevaluate the
potential impacts of the birth control policy.
Furthermore, my paper is also related to the literature on the causes of the dramatic demographic and social changes of the late 1960s and 1970s in the U.S. (“power of the pill” or “power
of abortion”). Goldin and Katz (2002) use a differences-in-differences specification to provide
evidence that female college graduates who legally could consent to the pill prior to age 18 married later and were more likely to pursue graduate education. Subsequent work has extended
this framework to produce additional evidence on the “power of the pill” in causing reduced
fertility (Bailey, 2006, 2009; Ananat and Hungerman, 2008), improved long-run outcomes for
children (Ananat and Hungerman, 2008) and other influences. However, using a new panel of
data on state policies related to access to the pill and abortion, Myers (2012) finds that access to
abortion had substantial effects on the probabilities of entering into marriage and motherhood.
On the other hand, the introduction of the pill and legal changes that granted young unmarried
women capacity to consent to it had little if any effect on the average probabilities of marrying
and giving birth at a young age. Finally, Pop-Eleches (2006) studies the abortion ban introduced by Romanian dictator Nicolae Ceausescu in 1966 and the impacts of the resulting higher
birth rates. Both Ceausescu’s Romania and China are communist regimes, and hence they can
introduce such dramatic policy restrictions toward individual fertility decisions.
8 For
example, Baige Zhao, the former vice minister of China’s National Population and Family Planning Commission (NPFPC), quoted this number in the Copenhagen U.N. climate conference in 2009. NHFPC was integrated
into National Health and Family Planning Commission (NHFPC) later.
8
3
3.1
Empirical Strategy
Regression Specifications
In OLS regressions, I will estimate the following equation:
yist = β1 + β2 SOEist + β3 Xist + λs + λt + εist ,
(1)
In this equation, i indicates the individual in province s and year t. The variable y represents
the number of children he has. SOE, the variable of major interest, indicates if this individual is
working in a SOE, X is a set of individual characteristics, including age, age squared, education
level, family income level, communist party affiliation and self-reported health, λs and λt are
province and year fixed effects.
An older individual will generally have more children since he is more likely to have finished
his lifetime fertility. I add a quadratic term to account for possible nonlinearity in age, for
example, the number of children might be smaller for the young people than a simple linear
relationship implies. Better educated people may receive less marginal benefit than marginal
cost from an extra child, and hence they may have less children than less educated people
(Becker, 1960). These people may also invest more resources on their children, and hence
reduce the number of children they can have. Becker et al. (2010) present evidence that such a
trade-off existed already in the nineteenth century. The negative relationship between education
and fertility is widely documented in different countries (Cleland, 1985). Basu (2002) review
mechanisms that education lead to lower fertility.
More wealthy parents have better ability to raise more children, but they are usually better
educated and probably receive less net benefit from children. Most empirical studies provide
negative impacts of family income (Borg, 1989). Communist party members are usually more
likely to be in SOEs with a correlation of 0.2 in my sample, so adding the party affiliation can
account for its independent impact on fertility. Health conditions may be important in fertility,
but the current self-reported health may be not the same as the health conditions when they were
in younger childbearing age.
The number of children varies from 0 to 7. The one-child policy might change the timing
people give birth. If people can give birth to two or more children, they might give the first
birth earlier and give another birth several years later. If they are allowed for having only one
child, they might delay the birth and in this way the total fertility rate will be lower as well.
This is the reason I also include the sampled individuals with no child. I also consider a setting
dropping individuals with no child, and the results are almost the same. Moreover, although the
9
most salient restriction imposed by the policy is restricting individuals having one child to give
another birth, the policy also restrict those already having two children to give the third birth.
In regressions with a dummy variable for one child or more than one child as the dependent
variable, I also find a strong negative impact of working in SOEs on the probability to have
more than one child.
3.2
Instrumental Variable and Identification
The classic selection problem probably exists in this study, because people working in a SOE
is not the same as those not. For example, their preference toward children and risk attitude
are probably different. If the individuals with a weaker preference toward children are more
likely to choose a job in a SOE, the negative relationship between the number of children and
the job type would be spurious. On the other hand, people working in SOEs are more likely
to prefer a stable life, and a family with more children might be more important for them.
Moreover, it is also possible that people with a higher ability or better political connection are
more likely to work in a SOE, while they also have better conditions to raise more children, or
can evade the penalty resulting from an extra birth. Therefore, the selection bias can be negative
or positive. Especially, since the two types of individuals may be quite different in unobservable
or unmeasurable factors, it is not feasible to just include a large set of covariates.
As a preliminary check about which kind of selection biases is dominating, I investigate if
there is a smaller desired number of children for never-married men working in a SOE than
those working in a non-SOE. There is a special question in this questionnaire: “Without policy
restrictions, how many children do you want to have”. I can obtain this variable for year 2010
and 2013. I keep those age less than or equal to 30 and obtain 147 observations. These people
usually just start their career and will probably marry later, so this gives us a chance to see if
individuals with a weaker preference toward children select into SOEs more likely. However,
people working in SOEs have a desired number of children of 1.63 on average, larger than 1.47,
the number for people working in non-SOEs actually. On the other hand, I compare the number
of children for ever-married men working in a SOE with those not for year 2010 and 2013
samples. The average number of children for SOE workers is 0.62, less than 0.65, the number
of non-SOE workers. Using the sampled individuals age less than or equal 35 give the similar
results. These results imply that positive selection, rather than negative selection, is probably
more important, and hence my OLS results probably underestimate the true impacts of working
in SOEs. Of course, it should be cautioned that these people are younger and hence are not
necessary the same as the older people in some dimensions.
Another problem involves job type switches, i.e. switch from a SOE job to a non-SOE job
10
or the converse. In general, job type switches are not common, especially for those spent most
of their careers during the planned economy era, and the first type of swithes is more common
than the second type because of the excessive employment in SOEs and the non-competitive
nature of China’s urban labor market. For example, Knight and Yueh (2009) and Wang et al.
(2014) document the SOE and non-SOE segmentation in China. More importantly, job type
swithes only lead to underestimation bias. For example, if a SOE worker loses his job at age 40
or older, he will be counted as a non-SOE worker in my regressions while he is not very likely
to give another birth at this age. However, during his prime childbearing ages, he was restricted
by the stricter policy in SOEs. A special case happens when an individual is fired by the SOE
he works because of giving an additional birth. However, this coincides my explanation for the
stronger restrictions imposed by SOEs actually. Intuitively, after switching from a SOE to a
non-SOE (though involuntarily), he can have more children.
To deal with the issues concerning about selection into a SOE and job type switches, I use
father’s SOE status as an instrumental variable. That is, I define a dummy variable, indicating
if his father was working in a SOE when he was 14 years old. Thus, I will estimate a first stage
in the following:
SOEist = δ1 + δ2 FSOEist + δ3 Xist + λs + λt + eist ,
(2)
where FSOE is father’s SOE status.
In the sample with father’s SOE status observed, there are only half of observations also
containing mother’s SOE status. Father is usually the primary earner in his family and has a
larger chance to affect his children’s career, hence I just use father’s SOE status rather than both
parents’, though regressions keep both parents’ SOE status as instrumental variables yield the
qualitatively identical results generally.
The excessive employment in SOEs is prominent in China. For example, after 18 years
of gradual transition, the SOE share in China’s total industrial output had declined from 77.6
percent in 1978 to 28.8 percent in 1996. However, in 1996 SOE’s still employed 57.4 percent
of urban workers and possessed 52.2 percent of total investment in industrial fixed assets (Lin
et al., 1998). Although a large-scale SOE reform in middle 1990s reduced the SOE share in
employment sharply, it is still generally acknowledged that in most SOEs the number of employees is larger than the profit maximization requires. Therefore, it is unusual for an individual
to enter a SOE if his first job is not a SOE job. Parents working in SOEs can probably use their
connections to assure a job for their children.
Moreover, the use of father’s SOE status is based on an institutional feature, which strongly
implies that father’s SOE status is plausibly exogenous. For a long period till the late 1990s,
children were allowed to take their parents’ jobs in SOEs one-by-one. For example, if both of
11
0
.01
.02
.03
.04
the parents are working in SOEs and they have two children, then both of the children can enter
the same firms upon their parents’ retirement or early retirement. Most people in my sample
began their career before 2000, hence this institutional feature is probably relevant for a large
share of individuals in my sample.
The major concern about this instrument is that father’s SOE status may affect his preference toward children and in turn affect his children’s directly, so that the exclusion restriction
is violated. To see why the exclusion constraint holds here, the special historical background
provides an important clue. Between the 1950s and 1970s, China was a planned economy, and
during the 1980s and 1990s, this country experienced a transition to a market economy. During
the planned economy era, citizens had little freedom to choose their jobs. The government usually distributed jobs based on its priority and funds availability. Figure 1 shows the distributions
of sampled individuals and their fathers’ birth years. Almost all the fathers were born before
1965, and hence probably began to work before 1990. This means the selection bias, even if
existing for many sampled individuals, is much less likely to be an issue for the group of fathers.
1900
1920
1940
1960
1980
2000
x
Birth year
Father's birth year
Figure 1: The distributions of sampled individuals and their fathers’ birth years
In the early 1950s when the communist party began to rule this country, private firms were
seized by the government. Since those firms were usually located in coastal areas, people in
those areas with many private firms nationalized were more likely to work in SOEs. Between
1953 and 1957 (the period for the “First Five-year Plan”), the government built 156 industrial
projects using the Soviet Union’s aid, and these projects were almost located in the Northeastern
part of this country (Manchuria) and provincial capitals. Not surprisingly, there was a large
demand for workers in SOEs during this period, and the government actually introduced a large
12
number of rural labor into SOEs and granted urban residence certificate (“Hukou”) to them.
1958-1961 witnessed a decline in the industrial production and then a recovery. Beginning with
the so called “Cultural Revolution” in 1966, the industrial production was hit seriously, and a
large number of urban graduates were sent down to rural areas, and it was hard for them to
find a SOE job even after they came back to the urban areas years later. In the early 1970s,
the government also built many industrial projects in the western and southern parts of this
country because the northern faced Soviet Union’s military threat. It hence can be expected that
people in these areas were more likely to find a job in SOEs. In the 1980s, the coastal areas
began to experience privatization and introduced foreign capital, and the share of SOE jobs was
relatively smaller there. Overall, no evidence shows that the government chooses a group of
workers with a special preference toward children. Almost all the fathers probably began their
career before 1990, and hence it is not very likely that people with a certain peference toward
children start working in SOEs. Starting from 1990s, the government no longer distributed jobs
to most people and the effective job markets emerged, then the citizens can choose their jobs
much more easily.
Is it possible for the preference toward children to be changed after working in SOEs many
years? Especially, it is possible that these fathers have a weaker preference toward children
after tolerating the stricter state control in SOEs, and this weaker preference might affect their
children’s fertility decisions. To check such possibility, I run regressions for the group with
age 40 to 60 in my sample with the desired number of children as the dependent variable.
The ages for this group are closer to the fathers’ in my sampled individuals, and the majority
of them have some exposure to the one-child policy during their prime childbearing ages. If
their preference toward children was not negatively affected by the policy, it is even less likely
for most fathers of the sampled individuals to be affected because they are usually older and
have less or no exposure to the one-child policy during their prime childbearing ages. The
unconditional regression and the conditional regression with a full set of covariates and fixed
effects give coefficients of 0.13 and 0.04, with p-values of 0.03 and 0.49. Therefore, it is hard
to say fathers working in SOEs have a weaker preference toward children and probably directly
lead to the lower fertility for their children.
Using father’s SOE status as the intrumental variable also alleviate the underestimation bias
resulting from job type switches. Overall, the fathers are much more likely to spend most of
their career in the planned economy era, when the government promised a lifetime job usually.
The 2SLS regressions give the estimates for the sampled individuals manipulated by their fathers’ SOE status, i.e. those working in a SOE if their fathers did and not working in a SOE if
their fathers did not (“complier” group), and hence are free of the bias from job type switches.
13
The underlying logic is that the current SOE status is related to whether the first job was in a
SOE or not, which is in turn related to father’s SOE status. Alleviating the underestimation bias
from job type switches probably explains why we obtain larger 2SLS results.
Furthermore, I consider two types of robustness check for the validity of the instrumental
variable. The strict one-child policy was introduced in 1979. Fathers born in or before 1940
had left their prime childbearing ages when the one-child policy was implemented, and hence
they were not very likely to be restricted by this policy and hence their preference was not
very likely to be affected because they were working in SOEs. I conduct a robustness check
using the sample of individuals whose fathers were born in or before 1940. The magnitude is
even larger, and the significance is lower but still significant under the 5% level. There may be
still some concerns about the “Wan, Xi, Shao” (later, longer, fewer) policy introduced in early
1970s, however, using the sample of individuals whose fathers were born in or before 1940, I
still obtain the similar results. In another check, I add father’s age, education level and party
member status in robustness check. These variables can account for the influence of fathers
on their children not through their children’s SOE status. For example, fathers with a higher
education level may prefer a lower number of children and affect their children’s willing to give
birth, while these fathers are probably more likely to work in a SOE. The results remain similar.
Finally, core family is nearly as common in urban China as in the Western countries, thus this
also limits their parents’ impact.
4
Data and Descriptive Statistics
My main data are obtained from the 2008, 2010 and 2013 samples of the China General Social
Survey (CGSS). CGSS is an annual or biannual survey of China’s urban and rural households
starting from 2003, carried out by the National Survey Research Center, Renmin University of
China (NSRC). It is generally considered as a survey with high quality, comparable with the
U.S. or Canada’s surveys with the same name. Using a multi-stage stratified sampling design,
this survey covers all the 31 provincial units in mainland China. The original samples for these
three years contain 6000, 11783 and 11438 observations respectively. Compared to the usual
datasets in economics, this survey is designed to gather data on social trends and the changing
relationship between social structure and quality of life in China. Although the survey was also
carried out in several other years, only the three samples contain the major variables used in my
study.
I keep the male sample with urban residence certificate (“Hukou”) age 18-60, married or
once married. Age 60 is the regular age a man can retire and collect retirement benefits. More14
over, none of the never-married individuals have children in the dataset, so I do not include
them in my sample. My sample contains 1767 observations after dropping those missing the
number of children, their own SOE status and their fathers’ SOE status or other individual characteristics, including age, education level, family income level, marital status, Ethnic Han, party
affiliation and self-reported health.
In China, the relatively loose control over rural residents is based on the residence certificate,
while the official one-child restriction is quite rigid across the citizens with urban residence
certificate, especially those in the same province. That is, although plenty of rural migrant
workers work in urban areas, they are not restricted by the stricter control over fertility since
they usually do not have urban residence certificate. I keep the male sample to avoid double
counting the impacts of SOE status for a family. Another major reason is that the husband is
usually the primary earner in a family, and hence job type should have a larger impact on him.
In the appendix, I show the negative impact of SOE status also exists for the female sample,
though the magnitude is smaller as expected, and in the OLS regressions husband’s SOE status
has a much stronger impact than her own SOE status.
The descriptive statistics can be seen in Table 19
9 Since
1980, the official minimum marriage ages for men and women are 22 and 20 respectively, i.e. the
minium age they can obtain the certificates of marriage. This is probably the reason why the minimum age in my
sample is 22. However, some individuals might marry earlier than the official minimum marriage ages unofficially,
and then obtain the certificates when reaching these ages. For this reason, I prefer to use the 18-60 as the range of
age, even if there is not a person age 18-21 in my sample.
15
Table 1: Descriptive Statistics
Variable
Observations Mean Std. Dev. Min Max
Panel A: Individual variables
No. of children
1767
1.09
0.58
0
7
No. of children (SOEs)
907
1.05
0.48
0
4
No. of children (non-SOEs)
860
1.13
0.67
0
7
Work in SOEs
1767
0.51
Age
1767
42.52
9.26
22
60
Log family income
1767
10.68
0.90
5.99 15.61
Education
not any
1767
0.01
primary school
1767
0.04
secondary school
1767
0.61
college
1767
0.35
Self-reported health
fair
1767
0.61
good
1767
0.33
very good
1767
0.06
Ethnic Han
1767
0.95
Party member
1767
0.25
Panel B: Father’s variables
Father in SOEs
1767
0.66
Father’s age
1641
72.18
11.87
43
110
Father’s party member
1767
0.29
Father’s education
not any
1749
0.16
primary school
1749
0.36
secondary school
1749
0.40
college
1749
0.08
Note: the urban male sample with age 18-60, married or once married, from China General Social Survey
(CGSS) 2008, 2010 and 2013 samples.
From Table 1, we can see that on average, the number of children is 1.09. This perhaps
implies a strong impact of the one-child policy. However, the standard deviation of the number
of children is 0.58, which means that although the policy may restrict the birth significantly, the
restrictions are varying in intensity across the country. Another reason is that many of sampled
individuals are still in their prime childbearing ages and hence their current number of children
is not equal to their total fertility. Investigating the differences across individuals working in
SOEs or not, I find that those working in SOEs have a smaller number of children on average,
and interestingly they also have a much smaller standard deviation.
To see if the differences in the number of children are concentrated in a certain group, I plot
the average number of children for SOE workers and non-SOE workers for eight birth cohorts
(adjusting for sampling years): 1948-1952, 1953-1957, 1958-1962, 1963-1967, 1968-1972,
1973-1977, 1978-1982, 1983-199110 . As Figure 2 shows, for all birth cohorts, SOE workers
10 We
have three sampling years spanning 5 years, hence the number of children for the same birth cohort in
16
.6
Average number of children
1
.8
1.2
1.4
1.6
have a smaller average number of children. The gap is smaller for younger cohorts, which is
not surprising because they are still in prime childbearing ages.
1948/1952 1953/1957 1958/1962 1963/1967 1968/1972 1973/1977 1978/1982
Birth cohort
Non-SOE workers
1983+
SOE workers
Figure 2: The average number of children for SOE workers and non-SOE workers for eight birth cohorts
The share of individuals working in a SOEs is less than the share of their fathers by 15
percentage points. This difference is large considering father’s SOE status asked in the questionnaire is when the individuals were 14 years old, and reflects the privatization of the economy
in China. For the individuals born after 1980, many of their parents experienced the large-scale
SOE reform and lost their jobs in a SOE in the late 1990s, and the share of their fathers working
in a SOE is 55%. For the individuals born in or before 1980, the share is 67%. The ratio of
individuals working in SOEs is much larger than the number reported by the National Bureau
of Statistics, but this is perhaps not quite surprising. There are 250 million rural migrant workers in urban areas in 2010, generally working in non-SOEs, but very few of them have urban
residence certificate, and hence are not the target group of my study. However, they are counted
by the National Bureau of Statistics as urban labor.
Although the number of individuals with college education is large, we may remember they
are from the urban sample. In China, the college attendance rate in urban areas is much higher
than in rural areas. The share of people with primary education is much less than this share for
their fathers, while the share of people with college education is much larger than the share for
different sampling year is not perfectly comparable without for adjusting sampling years. To adjust for sampling
years, I run regressions with the number of children as the dependent variable and SOE status as the explanatory
variable, including sampling year fixed effects fo the eight cohorts. I keep the estimated intercept as the average
number of children for non-SOE workers, and the difference between the estimated intercept and the coefficient
for SOE status as the average number of children for non-SOE workers.
17
their fathers, which is consistent with the general impression of a large-scale advancement in
educational attainment in this county.
5
5.1
Results
OLS Results
Table 2 presents the OLS results. I find a significantly negative relationship between working
in a SOE and the number of children in all settings, and the magnitudes are similar. In the
regression with all covariates, I find the average number of children for individuals working in
a SOE is 0.09 lower than those not working in a SOE. In other words, if the individuals not
working in SOEs came to SOEs, we would probably lose 10 children for each one hundred of
them. Concerning the large population size in China, this implies a large impact of privatization.
As a remainder, there is a selection issue in job type, and the direction of the selection bias is
also ambiguous.
18
Table 2: OLS regressions of the number of children on SOE status
Dep. variable: No. of children
Work in SOEs
(1)
-0.0761**
(0.0349)
(2)
-0.115***
(0.0314)
0.0844***
(0.0146)
-0.000728***
(0.000168)
(3)
-0.0840***
(0.0257)
0.0831***
(0.0144)
-0.000741***
(0.000168)
0.575***
(0.150)
0.238*
(0.127)
0.152
(0.128)
(4)
-0.0887***
(0.0257)
Age
0.0839***
(0.0146)
Age squared
-0.000747***
(0.000171)
Primary education
0.558***
(0.141)
Secondary education
0.209*
(0.114)
College education
0.102
(0.120)
Log family income
0.0144
(0.0298)
Ethnic Han
-0.192
(0.162)
Party member
0.0444
(0.0303)
Good health
-0.00471
(0.0704)
Very good health
0.060
(0.0721)
Province FE
Y
Y
Y
Year FE
Y
Y
Y
Observations
1767
1767
1767
1767
Adj. R-squred
0.004
0.211
0.227
0.229
Note: the excluded category of self-reported health is fair health, and the excluded category of education is not
any education. Standard errors are clustered at province,* p<0.10, ** p<0.05, *** p<0.01.
The number of children exhibits a strong positive relationship with respect to age, though
there is a slight quadratic pattern. This means that an older individual is more likely to have
achieved his lifetime fertility. The quadratic term of age is also significant so that the number
of children might be smaller for the young people than a simple linear relationship implies.
Although it seems natural to think that health condition has effects on fertility, I do not find a
significant impact of self-reported health. This is perhaps resulting from the fact that the current
health condition does not reflect the real health condition when an individual was giving birth
effectively, and the health condition when an individual was giving birth was usually good.
Education level and the number of children exhibits a negative relationship, supporting the
theory of increasing net marginal cost of children for well-educated people. Party members
have a larger number of children, perhaps resulting from better economic conditions or a higher
ability to evade policy restrictions, but this estimate is not precise.
19
5.2
2SLS results
The first stage of my 2SLS regressions in Table 3 shows a significantly positive impact of
fathers’ SOE status on their children’s. The F statistic of father’s job type in the regression
including all covariates and fixed effects is 34.52, much larger than 10, the usual critical value
for judging weak instruments (Stock and Yogo, 2005), and hence supports the validity of my
intrumental variable. To address the concern about functional form mispecifications, I also run
Probit and Logit regressions. The average marginal effects from these regressions are very
similar to those from OLS regressions and are highly significant as well. Because in 2SLS
regressions it is not appropriate to use a nonlinear model in the first stage, I do not report the
Probit and Logit results for simplicity.
Moreover, older people are more likely to work in SOEs, reflecting the privatization of the
economy in last three decades. Party members are more likely to be seen in SOEs as expected.
Moreover, although better educated people are more likely to work in SOEs, the impact is not
significant.
20
Table 3: 2SLS regressions: first stage
Dep. variable: Work in SOEs
Father in SOEs
(1)
0.257***
(0.0372)
(2)
0.220***
(0.0375)
0.0257
(0.0168)
-0.000247
(0.000193)
(3)
0.200***
(0.0348)
0.0293*
(0.0156)
-0.000246
(0.000181)
-0.153
(0.193)
0.0917
(0.180)
0.278
(0.187)
(4)
0.200***
(0.0340)
Age
0.0264
(0.0157)
Age squared
-0.000236
(0.000184)
Primary education
-0.169
(0.199)
Secondary education
0.0386
(0.191)
College education
0.166
(0.202)
Log family income
0.00386
(0.0176)
Ethnic Han
-0.00824
(0.0479)
Party member
0.173***
(0.0355)
Good health
0.0512
(0.0522)
Very good health
0.0209
(0.0584)
Province FE
Y
Y
Y
Year FE
Y
Y
Y
F-stat for excluded instrument
47.84
34.48
32.99
34.51
Observations
1767
1767
1767
1767
Adj. R-squred
0.059
0.097
0.134
0.151
Note: the excluded category of self-reported health is fair health, and the excluded category of education is not
any education. Standard errors are clustered at province,* p<0.10, ** p<0.05, *** p<0.01.
My 2SLS results show a much larger impact of privatization. In the regression with all
covariates, I find the average number of children for individuals working in a SOE is 0.42 lower
than those not working in a SOE. That is, without considering the selection issue and job type
switches, we may underestimate the impact of SOE status by a large amount. Of course, the
2SLS estimates are specific to a “complier” group as usual, and their reaction to SOE status
may be not equal to the average person. The magnitudes of other covariates do not change very
much compared to OLS results, though the standard errors are larger as usually seen in 2SLS
regressions. Hence the same explanations in Section 5.1 apply here.
As explained in Section 3.1, people working in SOEs are more likely to prefer a stable life,
and a family with more children might be more important for them. Moreover, it is also possible
that people with a higher ability or better political connection are more likely to work in a SOE,
while they also have better conditions to raise more children, or can evade the penalty resulting
from an extra birth. Although there probably exists negative selection bias as well, this positive
21
selection bias seems dominating, and using father’s SOE status as the instrumental variable alleviate these selection biases. Moreover, although the number of job type switches is probably
small, this small number of job type switches can probably lead to large underestimation. Overall, the fathers are much more likely to spend most of their career in the planned economy era,
when the government ensured a lifetime job usually. The 2SLS regressions give the estimates
for the sampled individuals manipulated by their fathers’ SOE status, i.e. those working in a
SOE if their fathers did and not working in a SOE if their fathers did not (“complier” group),
and hence are free of the bias from job type switches. These facts probably explains why we
obtain larger 2SLS results.
Table 4: 2SLS regressions: second stage
Dep. variable: No. of children
Work in SOEs
(4)
-0.414**
(0.205)
Age
0.0949***
(0.0203)
Age squared
-0.000850***
(0.00023)
Primary education
0.505***
(0.164)
Secondary education
0.247
(0.160)
College education
0.181
(0.174)
Log family income
0.0191
(0.0273)
Ethnic Han
-0.195
(0.160)
Party member
0.0999*
(0.0513)
Good health
0.0103
(0.0699)
Very good health
0.0633
(0.0707)
Province FE
Y
Y
Y
Year FE
Y
Y
Y
Observations
1767
1767
1767
1767
Note: the excluded category of self-reported health is fair health, and the excluded category of education is not
any education. Standard errors are clustered at province,* p<0.10, ** p<0.05, *** p<0.01.
6
(1)
-0.537***
(0.181)
(2)
-0.479**
(0.196)
0.0968***
(0.0208)
-0.000852***
(0.000234)
(3)
-0.420**
(0.206)
0.0954***
(0.0204)
-0.000849***
(0.000228)
0.527***
(0.176)
0.297*
(0.170)
0.274
(0.186)
Robustness Check
Alternative setting: one child or more than one child To see the probably most salient impact
of the one-child policy, i.e. restricting those having one child to give another birth, I define a
22
dummy variable, which is equal to one if an individual has more than one child, and zero if he
has only one child. The share of SOE workers having more than one child is 0.13, while this
share for non-SOE workers is 0.19. This difference is significant under the 5 percent level.
I replace the number of children with this dummy variable as the dependent variable and
run Probit and IV probit regressions. Using Logit and OLS regressions and their IV versions
give the similar results. I find a strong negative impact of working in SOEs on the probability
to have more than one child, as we can see from the average marginal effects in Table 5. The
results of Probit and IV Probit with a full set of covariates and fixed effects show that compared
to those not working in SOEs, people working in SOEs are 5.7 percent and 34.4 percent less
likely to have more than one child. The reasons why we obtain larger IV results are similar as
before. Probably not surprising, the estimates of other covariates are also almost the same in
direction and significance as before.
Table 5: Probit regressions of the dummy variable for more than one child
Dep. variable:
More than one child
Work in SOEs
Probit
(1)
-0.057***
(0.017)
IV Probit
(4)
-0.346***
(0.103)
Age
0.029**
(0.013)
Age squared
-0.0002
(0.0001)
Primary education
0.154*
(0.092)
Secondary education
0.119
(0.095)
College education
0.106
(0.108)
Log family income
0.008
(0.017)
Ethnic Han
-0.080*
(0.041)
Party member
0.092***
(0.026)
Good health
0.021
(0.034)
Very good health
0.027
(0.037)
Province FE
Y
Y
Y
Year FE
Y
Y
Y
1st-stage F for excluded instrument
33.82
28.06
Observations
1599
1599
1599
1599
Note: the excluded category of self-reported health is fair health, and the excluded category of education is not
any education. Standard errors are clustered at province,* p<0.10, ** p<0.05, *** p<0.01.
23
Probit
(2)
-0.061***
(0.017)
0.019**
(0.010)
-0.0001
(0.0001)
0.256***
(0.069)
0.094
(0.064)
0.041
(0.071)
0.006
(0.019)
-0.083*
(0.044)
0.048**
(0.023)
0.009
(0.036)
0.027
(0.038)
Y
Y
IV Probit
(3)
-0.328***
(0.089)
Additional check for the validity of the instrumental variable To address the concern that
father’s SOE status may affect their preference toward children and in turn affect their children’s
directly, I consider a robustness check using the sample of individuals whose fathers were born
in or before 1940. The strict one-child policy was introduced in 1979. Fathers born in or before
1940 had left their prime childbearing ages when the one-child policy was implemented, and
hence they were not very likely to be restricted by this policy and hence their preference was
not very likely to be affected because they were working in SOEs. As shown in the first column
in Table 6. The magnitude is even larger, and the significance is lower but still significant under
the 5% level.
There may be still some concerns about the “Wan, Xi, Shao” (later, longer, fewer) policy
introduced in early 1970s, however, using the sample of individuals whose fathers were born in
or before 1940, I still obtain the similar results. Although because of the smaller sample size,
the standard error is larger.
More father’s variables In a separate robustness check, I add father’s age, education level
and party member status in 2SLS regressions. As mentioned earlier, these variables can account
for the influence of parents on their children not through their children’s job type. For example,
parents with a higher education level may prefer a lower number of children and affect their
children’s willing to give birth, while these parents are probably more likely to work in a SOE.
As shown in column 3 in Table 6, the results remain similar.
Although it is tempting to include wife’s SOE status in the regressions as well, it is unlikely
for wife’s SOE status to be exogenous. An individual tends to find someone similar to him/her
in education level, income level, and social class according to the assortative mating theory.
Indeed, their SOE status are highly correlated, with a correlation coefficient of 0.42 in my
sample. In addition, I find husband’s SOE status has a much stronger impact in the linear
regressions for the women sample as shown in the Appendix.
24
Table 6: More robustness check
Dep. variable:
No. of children
Work in SOEs
More father’s variables
(3)
-0.424**
(0.202)
Other covariates
Y
Other father’s variables
Y
Province FE
Y
Y
Y
Year FE
Y
Y
Y
1st-stage F for excluded instrument
22.07
8.58
37.20
Observations
971
484
1629
Note: the first and second columns are 2SLS results for individuals whose fathers were born in or before
1940,and individuals whose fathers were born in or before 1930, and the third column is 2SLS results when
adding father’s age, education level and party affiliation, with parents’ job types as instruments. Other covariates
are age, age squared, education level, family income level, Ethnic Han, party affiliation and self-reported health.
The excluded category of self-reported health is fair health, and the excluded category of education is not any
education. Standard errors are clustered at province,* p<0.10, ** p<0.05, *** p<0.01.
7
Father Born in/before 1940
(1)
-0.681**
(0.299)
Y
Father born in/before 1930
(2)
-0.691*
(0.413)
Y
Conclusion
This paper evaluates the role of China’s privatization of the economy in weakening the effects
of the birth control policy, known as the one-child policy in the West, during the last three
decades. Working in a SOE implies more control over individual choices by the government.
The baseline OLS regressions show that compared to those not working in a SOE, the number
of children for individuals working in a SOE is about 0.1 less on average. In other words,
if the individuals not working in SOEs came to SOEs, we would probably lose 10 children
for each one hundred of them. To deal with the selection issue, I use father’s SOE status as
an instrumental variable. The 2SLS results show a much larger difference in the number of
children: 0.4. In regressions with a dummy variable for one child or more than one child as the
dependent variable, I also find a strong negative impact of working in SOEs on the probability
to have more than one child.
My results have a strong policy implication. China’s total fertility has been lower than the
replacement level for two decades, with a large share of aged population. Although the onechild policy was originally designed to be an one-generation policy (Fong, 2006), the policy is
only relaxed to a mild extent till recently. My results imply that the birth control policy might
already lead to an even lower fertility rate in China without privatization of the economy. And
since the share of employment in SOEs is already low, the potential of privatization in further
weakening the birth control policy is also limited. This implies that a further adjustment or
abolition of this policy is needed. Considering China’s low fertility rate compared to the world
average, it is probably not surprising for the government to encourage giving birth one day. The
25
increased state capacity in SOEs is probably useful in implementing this potential policy.
26
References
[1] Acemoglu, Daron, Camilo Garcia-Jimeno and James A. Robinson (2015). “State Capacity and Economic Development: A Network Approach,” American Economic Review,
105(88), pp. 2364-2409.
[2] Amsden, Alice H. (1989). “Asia’s Next Giant: South Korea and Late Industrialization,”
New York: Oxford University Press.
[3] Ananat, Elizabeth Oltmans and Daniel Hungerman (2008). “The Power of the Pill for the
Marginal Child: Oral Contraception’s Effects on Fertility, Abortion, and Maternal and
Child Characteristics, 2008,” NBER working paper.
[4] Bailey, Martha (2006). “More Power to the Pill,” Quarterly Journal of Economics, 121(1),
pp. 289320.
[5] Bailey, Martha (2009). “More Power to the Pill: Errata and Addendum,” Quarterly Journal
of Economics, Published online February 17, 2010.
[6] Bandyopadhyay, Sanghamitra, and Elliot D. Green (2012). “Pre-Colonial Political Centralization and Contemporary Development in Uganda,” Paper presented at the American
Political Science Association Annual Meeting, New Orleans, LA.
[7] Basu, Alaka Malwade (2002). “Why Does Education Lead to Lower Fertility? A Critical
Review of Some of the Possibilities,” World Development, 30(10), pp. 17791790.
[8] Becker, Gary (1960). “An Economic Analysis of Fertility,” in Demographic and Economic
Change in Developed Countries.
[9] Becker, Sascha O., Francesco Cinnirella and Ludger Woessmann (2010). “The Tradeoff between Fertility and Education: Evidence from before the Demographic Transition,”
Journal of Economic Growth, 15(3), pp 177-204.
[10] Bennett, John, Saul Estrin and Giovanni Urga (2007). “Methods of Privatization and Economic Growth in Transition Economies,” Economics of Transition, 15(4), pp. 661-683.
[11] Bonin, John P., Iftekhar Hasan and Paul Wachtel (2005). “Privatization Matters: Bank
Efficiency in Transition Countries,” Journal of Banking & Finance, 29(8), pp. 2155-2178.
[12] Borg, Mary OMalley (1989). “The Income-Fertility Relationship: Effect of the Net Price
of a Child,” Demography, 26(2), pp. 301-310.
27
[13] Brainerd, Elizabeth (2002). “Five Years after: The Impact of Mass Privatization on Wages
in Russia, 19931998,” Journal of Comparative Economics, 30(1), pp. 160-190.
[14] Cai, Yong (2010). “China’s Below-replacement Fertility: Government Policy or Socioeconomic Development?” Population and Development Review, 36(3), pp. 419-440.
[15] Centeno, Miguel A. (2002). “Blood and Debt: War and Nation-State in Latin America,”
Princeton: Princeton University Press.
[16] Chen, Jiajian, Robert D. Retherford, Minja Kim Choe, Xiru Li, and Ying Hu. (2009).
“Province Level Variation in the Achievement of Below-replacement Fertility in China,”
Asian Population Studies, 5(3), pp. 309-327.
[17] Cleland, John (1985). “Marital Fertility Decline in Developing Countries: Theories and
the Evidence,” in Reproductive Change in Developing Countries: Insights from the World
Fertility Survey, ed. by John Cleland and John Hobcraft, in collaboration with Betzy Dinesen. London, England, Oxford University Press, pp. 223-252.
[18] Estrin, Saul, Jan Hanousek, Evzen Kocenda and Jan Svejnar (2009). “The Effects of Privatization and Ownership in Transition Economies,” Journal of Economic Literature, 47(3),
pp. 699-728.
[19] Evans, Peter (1995). “Embedded Autonomy: States and Industrial Transformation,”
Princeton: Princeton University Press.
[20] Fong, Vanessa L. (2006). “Only Hope: Coming of Age under China’s One-child Policy,”
Stanford University Press.
[21] Frydman, Roman, Cheryl Gray, Marek Hessel and Andrzej Rapaczynski (1999). “When
Does Privatization Work? The Impact of Private Ownership on Corporate Performance in
the Transition Economies,” Quarterly Journal of Economics, 114(4), pp. 1153-1191.
[22] Gennaioli, Nicola and Ilia Rainer (2007). “The Modern Impact of Precolonial Centralization in Africa,” Journal of Economic Growth, 12(3), pp. 185234.
[23] Goldin, Claudia and Lawrence Katz (2002). “The Power of the Pill: Oral Contraceptives
and Womens Career and Marriage Decisions,” Journal of Political Economy, 110(4), pp.
730770.
28
[24] Gu, Baochang, Feng Wang, Zhigang Guo, and Erli Zhang. (2007). “China’s Local and National Fertility Policies at the End of the Twentieth Century,” Population and Development
Review, 33(1), pp.129-147.
[25] Herbst, Jeffrey (2000). “States and Power in Africa: Comparative Lessons in Authority
and Control,” Princeton: Princeton University Press.
[26] Johnson, Chalmers A. (1982). “MITI and the Japanese Miracle: The Growth of Industrial
Policy,” pp. 19251975, Stanford: Stanford University Press.
[27] Knight, John and Linda Yueh (2009). “Segmentation or Competition in China’s Urban
Labour Market?” Cambridge Journal of Economics, 33(1), pp. 79-94.
[28] La Porta, Rafael, Florencio Lopez-De-Silanes and Andrei Shleifer (2002). “Government
Ownership of Banks,”Journal of Finance, 57(2002), pp. 265301.
[29] Lastarria-Cornhiel, Susana. (1997). “Impact of Privatization on Gender and Property
Rights in Africa,” World Development, 25(8), pp.1317-1333.
[30] Lin, Justin Y., Fang Cai and Zhou Li (1998). “Competition, Policy Burdens, and Stateowned Enterprise Reform,” American Economic Review Papers and Proceedings, 88(2),
pp. 422-427.
[31] Michalopoulos, Stelios, and Elias Papaioannou (2013). “Pre-Colonial Ethnic Institutions
and Contemporary African Development,” Econometrica, 81(1), pp. 113152.
[32] Myers, Caitlin Knowles (2012). “Power of the Pill or Power of Abortion? Re-examining
the Effects of Young Women’s Access to Reproductive Controlt,” working paper.
[33] PopEleches, Cristian (2006). “The Impact of an Abortion Ban on Socioeconomic Outcomes of Children: Evidence from Romania,” Journal of Political Economy, 114(4), pp.
744-773.
[34] Poston, Dudley L. (2000). “Social and Economic Development and the Fertility Transitions in Mainland China and Taiwan,” Population and Development Review, 26(Supp.),
pp. 40-60.
[35] Poston, Dudley and Baochang Gu (1987). “Socioeconomic Development, Family planning, and Fertility in China,” Demography, 24(4), pp. 531-551.
29
[36] Stock, James and Motohiro Yogo (2005). “Testing for weak Instruments in linear IV regression,” in Identification and Inference for Econometric Models, ed. by Donald Andrews, New York: Cambridge University Press, pp. 80-108.
[37] Wade, Robert H. (1990). “Governing the Market,” Princeton: Princeton University Press.
[38] Wang, Haining, Fei Guo and Zhiming Cheng (2014). “A Distributional Analysis of Wage
Discrimination against Migrant Workers in China’s Urban Labour Market,” Urban Studies.
30
Appendix: the major results for the female sample
Table 7: The major results for women sample
Dep. variable:
No. of children
Work in SOEs
OLS
(1)
-0.0209
(0.0245)
2SLS
(4)
-0.803**
(0.395)
Husband in SOEs
0.186
(0.141)
Other covariates
Y
Y
Y
Province FE
Y
Y
Y
Year FE
Y
Y
Y
First-stage F-stat for excluded instrument
16.93
6.70
Observations
1542
1213
1542
1213
Note: the 2SLS regressions use father’s job type as an instrument. Other covariates are age, age squared,
education level, family income level, Ethnic Han, party affiliation and self-reported health. The excluded category
of self-reported health is very good health, and the excluded category of education is not any education. Standard
errors are clustered at province,* p<0.10, ** p<0.05, *** p<0.01.
31
OLS
(2)
-0.0225
(0.0245)
-0.0818***
(0.0292)
Y
Y
Y
2SLS
(3)
-0.387*
(0.210)