Close
About
FAQ
Home
Collections
Login
USC Login
Register
0
Selected
Invert selection
Deselect all
Deselect all
Click here to refresh results
Click here to refresh results
USC
/
Digital Library
/
University of Southern California Dissertations and Theses
/
Essays in development economics
(USC Thesis Other)
Essays in development economics
PDF
Download
Share
Open document
Flip pages
Contact Us
Contact Us
Copy asset link
Request this asset
Transcript (if available)
Content
ESSAYS IN DEVELOPMENT ECONOMICS
by
Xiongfei Li
A Dissertation Presented to the
FACULTY OF THE USC GRADUATE SCHOOL
UNIVERSITY OF SOUTHERN CALIFORNIA
In Partial Fulfillment of the
Requirements for the Degree
DOCTOR OF PHILOSOPHY
(ECONOMICS)
May 2023
Copyright 2023 Xiongfei Li
I dedicate this thesis to myself, my family, my friends, and to those who have
helped me become who I am now.
ii
Acknowledgements
I am profoundly grateful for the guidance and assistance of my advisors. John Strauss has
been an exemplary mentor, consistently offering invaluable insights and recommenda-
tions for both the overarching aspects and finer details of my research. Jeffrey Weaver has
been a constant source of encouragement and motivation while emphasizing the impor-
tance of rigor in my work. Daniel Bennett has influenced my perspective on experimental
development research and provided excellent career guidance. Neha Bairoliya’s expertise
in Macroeconomics has significantly expanded the scope of my research.
I would also like to express my appreciation for the invaluable support and advice pro-
vided by other faculty members at the University of Southern California, including but
not limited to, Vittorio Bassi, Augustin Bergeron, Nan Jia, Shelley Xin Li, Jeffrey Nuggent,
Paulina Oliva, Geert Ridder, Simon Quach, and Simone Schaner. Furthermore, I am grate-
ful to Na Zhao at Nankai University, and Laura Schechter, Jack Porter, and Chris Taber at
the University of Wisconsin-Madison for introducing me to the realm of rigorous eco-
nomic research.
The completion of this dissertation would not have been achievable without the un-
wavering support and care of my friends, peers, and family members, particularly Ruozi
Song and Rajat Kochhar. I would also like to extend my heartfelt gratitude to my parents
for their continuous support and encouragement.
iii
Contents
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Dedication . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ii
Acknowledgements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . iii
List of Tables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . vii
List of Figures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ix
Abstract . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . x
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
Chapter 1: The Unintended Impacts of the One-child Policy Relaxation in China
on Female Labor Market Outcomes . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.2 Background . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
1.3 Data Source . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
1.3.1 Sample Selection . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
1.3.2 Main Outcomes and Summary Statistics . . . . . . . . . . . . . . . . . 13
1.4 Empirical Strategy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
1.4.1 Static DID . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
1.4.2 Dynamic DID . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
1.4.3 Identification Assumptions . . . . . . . . . . . . . . . . . . . . . . . . 17
iv
1.4.4 Issues On Staggered TWFE Model . . . . . . . . . . . . . . . . . . . . 19
1.5 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
1.5.1 Impacts of Two-child Policy on New Birth . . . . . . . . . . . . . . . . 20
1.5.2 Impacts of Two-child Policy on Female Labor Market Outcomes . . . 21
1.5.3 Heterogeneous Impacts of Two-child Policy by Child Amount . . . . 23
1.5.4 Heterogeneous Impacts of Two-child Policy by Gender of First Child 25
1.5.5 Robustness Check . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
1.5.5.1 Results Using Alternative Estimation Methods . . . . . . . 26
1.5.5.2 Other Robustness checks . . . . . . . . . . . . . . . . . . . . 28
1.5.6 Dynamic Effects and Parallel Trends . . . . . . . . . . . . . . . . . . . 29
1.6 Mechanisms of Heterogeneous Effects on One-child Policy Compliers . . . . 32
1.7 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
Chapter 2: Income Inequality and Heterogeneous Return to Education in China . 40
2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
2.2 Background . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
2.2.1 Evolution of Income Inequality in China after 1978 . . . . . . . . . . 46
2.2.2 Rural-Urban Gaps and Income Sources . . . . . . . . . . . . . . . . . 48
2.3 Data Source . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52
2.4 Empirical Strategy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56
2.5 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58
2.5.1 Evolution of Education Returns in China . . . . . . . . . . . . . . . . 58
2.5.2 Distribution of the Education Return by Rural/Urban Divide . . . . . 60
2.5.3 Distribution of the Education Return by Highest Education Attained 60
2.5.4 Distribution of the Education Return by Gender . . . . . . . . . . . . 63
2.6 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63
Chapter 3: High-Speed Network and Self-Employment —Evidence from CFPS . 67
v
3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67
3.2 Background and Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 70
3.3 Empirical Strategy and identification challenge . . . . . . . . . . . . . . . . . 76
3.4 Main Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 78
3.4.1 Impacts on Family Self-employment Choice . . . . . . . . . . . . . . . 78
3.4.2 Impacts on Self-employment Income and Assets . . . . . . . . . . . . 79
3.4.3 Impacts on Individual Self-employment Choice . . . . . . . . . . . . 81
3.4.4 Heterogeneity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 82
3.5 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 84
Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 87
Appendices . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 103
F Empirical Strategy in Lemieux 2006b . . . . . . . . . . . . . . . . . . . . . . . 103
vi
List of Tables
1.1 Summary Statistics for Fertile-aged Females . . . . . . . . . . . . . . . . . . . 15
1.2 Regression Results on New Birth of All Fertile Aged Females and by Three
Age Groups . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
1.3 Regression Results on Labor Market Outcomes . . . . . . . . . . . . . . . . . 22
1.4 Regression Results on Female Labor Market Outcomes by Age Groups . . . 23
1.5 Heterogeneous Impacts by Child Amount in 2010 . . . . . . . . . . . . . . . 24
1.6 Heterogeneous Impacts by Gender of the First Child in 2010 . . . . . . . . . 26
1.7 Treatment Effects on Males and Elder Females . . . . . . . . . . . . . . . . . 29
1.8 Heterogeneous Impacts by Education Attainment in 2010 . . . . . . . . . . . 33
1.9 Heterogeneous Impacts by Community or Hukou Category in 2010 . . . . . 35
1.10 Heterogeneous Impacts by Occupation in 2010 . . . . . . . . . . . . . . . . . 36
1.11 Heterogeneous Impacts by Average Family Income in 2010 . . . . . . . . . . 37
2.1 Summary Statistics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55
2.2 Estimates of the Distribution of the Education Return . . . . . . . . . . . . . 59
2.3 Estimates of the Distribution of the Education Return by Rural/Urban Divide 61
2.4 Estimates of the Distribution of the Education Return by Highest Education
Attained . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62
2.5 Estimates of the Distribution of the Education Return by Gender . . . . . . . 64
3.1 Summary Statistics–Households . . . . . . . . . . . . . . . . . . . . . . . . . 73
3.2 Summary Statistics–Individuals . . . . . . . . . . . . . . . . . . . . . . . . . . 74
3.3 Pre-trend Check in Province Level . . . . . . . . . . . . . . . . . . . . . . . . 77
3.4 Internet Speed and Family Self-Employment Choice . . . . . . . . . . . . . . 78
vii
3.5 Internet Speed and Family Self-Employment Income and Assets . . . . . . . 80
3.6 Internet Speed and Individual Self-Employment Choice . . . . . . . . . . . . 81
3.7 Heterogeneity in Impacts of Individual Self-Employment Choice . . . . . . . 83
viii
List of Figures
1.1 Dynamic Impacts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
2.1 GDP per capita and Gini Index in China Since 1978 . . . . . . . . . . . . . . 41
2.2 Gini Index in China Since 1978 from Various Sources from 2002 to 2015 . . . 48
2.3 Urban-to-rural Income Ratio and Gini Index in China Since 1978 . . . . . . . 50
2.4 Rural and Urban Household Income Sources in China from 1998 to 2022 . . 51
ix
Abstract
This dissertation encompasses three chapters exploring development economics in China,
focusing on fertility’s impact on female labor market outcomes, income inequality trends,
and the relationship between technology and self-employment decisions.
In the first chapter, I use China’s national population policy shock as a natural exper-
iment to analyze fertility’s broader implications, including fertility expectations. Using
China Family Panel Survey (CFPS) data from 2010 to 2018 and a staggered difference in
differences (DID) research design, the findings demonstrate that affected couples are 40%
more likely to have a new birth, while eligible females experience larger declines in labor
participation, work hours, and salary earnings. The most pronounced declines are ob-
served among prime-age, better-educated urban females.
The second chapter examines income inequality in China over the past 30 years and as-
sesses changes in education returns distribution. By employing a new categorical random
coefficient model and data from the China Household Income Project (CHIP), the study
reveals a 54% decrease in overall education return in the recent decade, with increased
dispersion across rural/urban, education attainment, and gender divides. These findings
highlight potential future divergence in wage and household income.
In the third chapter, the correlation between technological advancements and self-
employment choices in China is investigated. Utilizing quarterly provincial internet speed
data and CFPS survey data, the analysis indicates that increased internet speed is asso-
ciated with a higher likelihood of self-employment, particularly among females, lower-
educated, and rural individuals. However, no evidence of a correlation with business
assets or income is found.
x
Introduction
This dissertation encompasses three chapters exploring development economics in China,
focusing on fertility’s impact on female labor market outcomes, education returns and
income inequality trends, and the relationship between technology and self-employment
decisions.
Wage gender gaps, where women earn less than men even when accounting for educa-
tion, skills, and experience, remain a significant issue in both developed and developing
nations. One major contributor to these persistent gaps is the unequal responsibility of
childcare. In the first chapter, ”Unforeseen Consequences of China’s One-Child Policy Re-
laxation on Women’s Labor Market Outcomes,” I leverage the shock of China’s national
population policy change as a natural experiment to examine the wider implications of
fertility, including fertility expectations, on women’s labor market outcomes. By analyz-
ing data from the China Family Panel Survey (CFPS) between 2010 and 2018, and using a
staggered difference in differences (DID) research design, the results reveal that affected
couples are 40% more likely to have a new birth, while eligible females experience twice-
size declines in labor participation, work hours, and salary earnings. These declines are
most pronounced among prime-age women (26 to 35 years old in 2010) and those who
complied with the one-child policy prior to the change. A further triple difference anal-
ysis indicates that greater impacts on labor supply are associated with better education,
community types, occupations, and average family income.
In addition to gender inequalities in the labor market, general income inequality is
also a topic with both theoretical and policy importance. In the second chapter ”Income
1
Inequality and Heterogeneous Return to Education in China”, we examine income in-
equality in China over the past 30 years and assess changes in education returns distribu-
tion. By employing a new categorical random coefficient model and data from the China
Household Income Project (CHIP) in 1995, 2002, 2007, 2013, and 2018, we can show both
the evolution of mean education returns and changes in dispersions. The study reveals a
54% decrease in overall education return in the recent decade, with increased dispersion
across rural/urban, education attainment, and gender divides. These findings highlight
potential future divergence in wage and household income.
As technology infrastructure advances, self-employment has emerged as a viable op-
tion, particularly for disadvantaged populations, alongside formal wage employment. In
the third chapter, ”High-Speed Internet and Self-Employment: Evidence from CFPS”, I in-
vestigate the relationship between technological progress and self-employment choices in
China. By collecting quarterly provincial internet speed data from the China Broadband
Development Alliance and CFPS survey data from 2012 to 2018, the analysis suggests that
higher internet speeds correlate with an increased likelihood of self-employment, espe-
cially among women, those with lower education levels, and rural residents. However, no
evidence of a relationship with business assets or income is found.
2
Chapter 1
The Unintended Impacts of the One-child Policy
Relaxation in China on Female Labor Market Outcomes
1.1 Introduction
Gender inequalities in labor market outcomes have been converging dramatically dur-
ing the past several decades. This “quiet revolution” is largely a result of the increasing
women’s labor participation and relative earnings (Goldin 2006). However, the conver-
gence process has largely stalled since the 1990s and the gender gaps are still prominent in
both developed and developing countries, especially among higher-skilled females (Blau
and Kahn 2017). After the attempts of anti-discrimination, job protection, and education
inclusion policies, the remaining gender inequalities can mostly be accounted for by un-
equal roles of parenthood for men and women (Kleven and Landais 2017; Kleven, Landais,
Posch, Steinhauer, and Zweim¨ uller 2019 and Cort´ es and Jessica Pan 2020).
Previous empirical research on the scale and causes of “child penalty” dominantly take
advantage of exogenous shocks to childbearing and child amount
1
. In this paper, I exploit
a unique policy change, China’s one-child policy relaxation as a shock to the household
1
Mark R Rosenzweig and Wolpin 1980 use the birth of twins as a shock to child amount; Angrist and
Evans 1998 use sex composition of first two children as an exogenous shock to the probability of having
the third child; Lundborg, Plug, and Rasmussen 2017 use in vitro fertilization (IVF) treatment success as a
variation to the birth of the first child; Markussen and Strøm 2022 use miscarriage us a shock to child amount
to study the impact of the first, second and the third childbirth.
3
fertility decision, to study how families react to the policy change and what further impacts
this policy shock has on female labor market outcomes. Different from the causal inference
designs of previous papers, I use timing variation in the eligibility of having a second child
among different couples as the identification strategy. Such context can provide not only
suggestive evidence of the direct impacts of childbirth after the policy shock but also bring
insights into the household decision of fertility and labor supply.
Population policies are common around the world, either encouraging or lowering fer-
tility rates. United Nations Department of Economic and Social Affairs, Population Divi-
sion 2021 documents that through 2019, there were nearly three-quarters of governments
globally had fertility policies (143 out of 197), of which 55 governments had policies to
raise fertility rates. The Chinese context is well suited to study family’s response to govern-
ment fertility policy change and its further impacts both because of the large population
affected and the extent to which the family policy has changed. China was not the only
country carrying out compulsory population control
2
, but it stands out for the persistence
of its strict one-child policy. The relaxation of the one-child policy restriction in China pro-
vides a natural experiment to study the general impacts of the fertility-encouraging policy
shock, concerning the differential response of families, and the role fertility penalties play
in this decision procedure.
The famous and controversial one-child policy in China was first implemented in late
1979. In 2013, the Chinese government eased the policy for limited couples if one of the
parents was a single child. In 2015, the long-term one-child policy was replaced by a
universal two-child policy, allowing all couples to have two children. Then in 2021, out of
concerns about sustained low fertility rates and the aging population, the fertility policy in
China was further revised to a three-child policy. In this paper, I focus on the policy change
in 2013 and 2015. The policy evolution process makes it possible to causally identify the
impacts of the policy change on fertility, and on female labor market outcomes, including
2
India during 1976 to 1977, Peru between 1996 to 2000 targeting indigenous population are two notice-
able examples besides China.
4
labor supply, weekly work hours, and salary earnings, using a staggered Difference in
Differences (DID) design.
In this paper, I use a national panel survey dataset, China Family Panel Survey(CFPS).
This family survey interviewed all family members from over 14,000 families in 2010 and
continued to do follow-up surveys every other year since 2012. I use 5 waves of surveys,
from 2010 to 2018, on which two waves are before the first policy change in 2013, one wave
between two policy changes, and two after the announcement of the universal two-child
policy. This dataset contains both demographic information and economic information, so
I can identify the eligibility of having a second child for each couple based on regulations in
each province. Using time variation among individual couples/females, I show that after
the policy was relaxed, 2.5% more eligible fertile-age females had a new birth between two
survey waves, while the percentage of females who were working around the survey time
also declined by 2.6 pp, and average weekly work time declined by 1.807 hours.
The heterogeneous analysis also shows that females in three age groups, 16 to 26 in
2010, 26 to 36 in 2010, and 36 to 45 in 2010 all responded to the policy relaxation and
had more childbirth, but the drop in labor supply only showed up among the prime-age
females, who were between 26 to 35 in 2010. Furthermore, heterogeneous analysis by child
amount in 2010 shows that females who had no child in the baseline year were less likely to
have childbirth after the policy relaxed, and females who had one child or more had more
new birth. The rise in the birth rate among females who had just one child in 2010, the
one-child policy compliers, is lower than the rise among females who already had two or
more children in 2010, and the labor supply decline only happened among the one-child
policy complying females.
These differential responses to the one-child policy relaxation can be rationalized by
differential price effects and income effects in the household decision of fertility. The
potential fine could be one reason why the one-child policy complying families/females
chose to have only one child. Higher opportunity costs and economic reasons could be
5
prior concerns in their fertility decision. Data show that the one-child policy complying
females were better educated, more likely to live in urban communities or own a non-
agricultural hukou, from families with higher average net income, and more likely to have
a non-agricultural job in the baseline year. These attributes mean that the opportunity cost
of having one more child is higher due to higher wage loss and less job flexibility, and the
family economic support is higher while support for child-caring is less.
Recent methodological papers point out that the TWFE estimates in a DID design can
be seriously biased because of treatment effects heterogeneity among observations in dif-
ferent groups at different periods (reviewed by Roth, Sant’anna, Bilinski, and Poe 2022 and
Chaisemartin and D’haultfoeuille 2022). This “forbidden comparison” (Goodman-Bacon
2021) problem is most common among staggered adoption designs. As part of the robust-
ness check, I first test the potential negative weights using the test method proposed by
Chaisemartin and D’Haultfœuille 2020 and then use the heterogeneity robust estimation
method proposed by Callaway and Sant’Anna 2021 with a smaller size of observations to
show that the less precise results are still in the small direction of my main results. More
robustness checks also show that my results are free of contamination and are not caused
by potential confounders.
This paper contributes to the literature in three strands. First, the topic of fertility
penalty and gender inequality in the labor market, reviewed by, for example, Altonji and
Blank 1999, Bertrand 2011, Olivetti and Petrongolo 2016, Blau and Kahn 2017, Olivetti and
Petrongolo 2016, Bertrand 2020, and Cort´ es and Jessica Pan 2020. To causally identify the
impacts of childbirth on female labor supply and earnings, researchers have attempted in
two directions. First, use exogenous variation to childbirth or child amount and make the
comparison within a small group of individuals (Mark R Rosenzweig and Wolpin 1980,
Angrist and Evans 1998, Lundborg, Plug, and Rasmussen 2017, and Markussen and Strøm
2022). The second practice is to use an event study method to estimate the trajectories of
6
within couple earning gaps after the birth of the first child (Angelov, Johansson, and Lin-
dahl 2016, Kleven, Landais, Posch, Steinhauer, and Zweim¨ uller 2019, and Kleven, Landais,
and Søgaard 2019). Kim and Moser 2021 use the same technique to study the productivity
loss of female scientists after the first childbirth). Besides drops in labor supply and earn-
ings, empirical findings also show that parenthood is also correlated with mothers’ choice
on job formality (I. Berniell, L. Berniell, Mata, Edo, and Marchionni 2021), job flexibil-
ity (Meekes and Hassink 2022), and overtime work hours (Cort´ es and Jessica Pan 2019).
Other than those, different impacts of grandparents’ distance (Akyol and Yilmaz 2021),
grandmother’s death (Talamas 2022), and closure of child facilities and schools during
the COVID-19 pandemic on mothers and fathers also result in unequal burden of child-
caring. To explain what causes the child penalty, one practice is to compare child penal-
ties among heterosexual non-adopting couples, adopting couples, and same-sex couples
(Kleven, Landais, and Søgaard 2021, Andresen and Nix 2022 and Levendis and Lowen
2022), and results suggest that the child penalty is mainly caused by preferences, gender
norms, and discrimination. In this paper, I exploit a unique policy shock, China’s one-
child policy relaxation, to study the impacts of this policy change on household fertility
decisions and female labor market outcomes. By doing regression with general fertile-age
females, I can show both suggestive evidence on child penalty after childbirth, and also
impacts on the general population, especially among those females who have not yet had
a new birth but had higher fertility expectations.
Second, to the impacts of government family policies. Government policies in female
employment protection and fertility encouragement are very common in the world, mainly
concentrated among high-income countries
3
. Empirical research on the actual effects of
such policies is still limited. Kleven, Landais, and Søgaard 2021 explored the impacts of
parental leave expansion and childcare subsidies since the 1950s in Austria and found no
3
United Nations Department of Economic and Social Affairs, Population Division 2021 indicates that
between 2015 and 2019, 28% of governments globally have policies to raise fertility, 58.7% governments in
Europe and North America, and 43.8% governments in Eastern and South-Eastern Asia have such policies.
7
contribution to the convergence of gender gaps. Ginja, Karimi, and P. Xiao 2023 study
similar policy reform in Sweden and show that the expansion of parental leave leads to
fewer recruitment and lower starting wages for young women, but not for men and el-
der women. Closer to my paper, Brenøe, Canaan, Harmon, and Royer 2020 study costs
of the female employee taking maternity leave on her firms and coworkers, and find out
negligible impacts. In this paper, I study a larger-scale national fertility policy change in
China and show evidence of differential responses from different families and suggestive
evidence of more general impacts.
Third, the empirical analysis of China’s universal two-child policy. The one-child pol-
icy in China has long been an attractive research topic (reviewed by J. Zhang 2017). Em-
pirical research on one-child policy has focused on fertility (Hongbin Li, J. Zhang, and
Zhu 2005 and Y. Cai 2010), child educational attainment (Hongbin Li, J. Zhang, and Zhu
2008, Mark R. Rosenzweig and J. Zhang 2009, B. Li and H. Zhang 2017, and Huang, X. Lei,
and A. Sun 2021), family distortion (Huang, X. Lei, and Y. Zhao 2016, Huang and Yi Zhou
2015), parental labor supply (F. Wang, L. Zhao, and Z. Zhao 2017) and sex ratio (Ebenstein
2010, Hongbin Li, J. Yi, and J. Zhang 2011). After more than 35 years, the one-child policy
was completely replaced by a universal two-child policy in 2015. This national essential
policy shock might have impacts on a broad scale and it is of both theoretical and policy
importance to evaluate this policy shock. Some researchers use two-child policy experi-
mentation in one county from 1985 to get insights into the potential effects of a universal
two-child policy (Qin and F. Wang 2017). The closest research to this paper is a fictitious
resume experiment conducted both before and after the policy change in 2015 (H. He,
S. X. Li, and Han 2023), they sent resumes in response to online job advertisements vary-
ing only on gender and only-child status. Callback results show that only-child females
received fewer calls than females with sibling(s) before 2016, and females with sibling(s)
got fewer calls than one-child counterparts after 2016. They also show that this discrimi-
nation against women is more pronounced among elder groups, which is consistent with
8
my findings that prime-age females experienced the highest drop in labor supply after the
policy changed. The advantage of my paper is that I use a family survey dataset covering
both before and after the policy change to evaluate the policy impacts, so I can get more
comprehensive results on how families/females respond to the policy relaxation and how
their labor market outcomes were affected. To the best of my knowledge, this paper pro-
vides the first thorough evaluation of the universal two-child policy in China, on fertility
decisions and female labor market outcomes.
The following part of this paper will be organized as followed: section two will be
the policy background; the third section will be empirical strategy and data sources; the
fourth part will be empirical results and robustness check; the fifth one will be mechanism
analysis, and the last section will be the conclusion.
1.2 Background
The population policies in China were capricious during the first twenty years after 1949.
Initially, the Chinese supreme leader Mao Zedong believed that more population ben-
efits social reform and economic development, and China experienced rapid population
growth in the early 1950s. After that, there were a few informal attempts at national family
planning policy after the first census result came out in 1953, and after the rapid popula-
tion growth rebound following the great famine from 1959 to 1961. However, these early
operations were all interrupted either by Mao’s ambitious Great Leap Forward movement
in 1958 or by the Cultural Revolution that began in 1966. A serious national population
planning campaign was conducted in 1971 after China’s total population exceeded 800
million in 1969. The initial slogan was “One child is not too few, two are just fine, and
three are too many”, and then the government announced another slogan “Later, Longer,
and Fewer”, which required late marriage (23 years old for females, and 25 for males), and
9
at most two children with more than a 3-year gap of married couples. This campaign suc-
cessfully decreased China’s overall fertility rate by half between 1971 and 1978 (J. Zhang
2017).
In 1979, one year after Deng Xiaoping gained leadership, a stricter population control
policy was enacted. The Chinese government started the compulsory one-child policy,
covering almost all provinces
4
in both rural and urban areas. After 1984, due to high re-
sistance, especially from rural families with only one daughter, the one-child policy started
to relax for some exceptional couples and stabilized until 2013 (F. Wang, L. Zhao, and Z.
Zhao 2017).
Before 2013, the one-child policy was strictly conducted for almost all urban couples
and most rural couples
5
, with exemptions that allowed a second child without fines as the
following:
• First, couples in which both the husband and the wife are only-child
• Second, Rural couples from five less populated and less developed provinces or au-
tonomous regions, including Hainan, Yunnan, Qinghai, Ningxia, and Xinjiang.
• Second, rural couples whose first child was a daughter, in all provinces except Shang-
hai (the so-called “one-and-a-half-child policy”)
6
.
• In Tianjin, Liaoning, Jilin, Shanghai, Jiangsu, Fujian, and Anhui, couples in which
just one of the wife or the husband was an only child.
• Couples in which both the husband and the wife belonged to a minority ethnicity.
4
Exemptions only exist in rural areas of Qinghai, Ningxia, Xinjiang, and Yunnan
5
In practice, the division between urban and rural was based on household registration status sys-
tem(hukou), which includes agricultural hukou (rural) and non-agricultural hukou (urban).
6
Besides six provinces or autonomous regions with more relaxed policies, Beijing, Tianjin, Chongqing,
Chengdu, and Jiangsu all had similar policies restricted to rural couples in mountain areas. In a table show-
ing different policies in different provinces/municipalities/autonomous regions for different populations,
categorizes these five provinces or municipalities into the ”one-and-a-half-child policy” provinces, and
Shanghai was the only municipality without this exemption.
10
Besides these exemptions, there were also more generous policies allowing three chil-
dren or unlimited children, which covered minority nomadic couples in some less popu-
lated and less developed areas. In Xizang, urban resident couples could have two children
in general and rural residents were not restricted by any family planning policy.
Besides policy variations, the actual policy implementation varies a lot in different
provinces or cities, in rural and urban areas, and for Han or minorities. Urban residents
working in state-owned companies may lose their jobs and pension if break the one-child
policy, while there was no such comparative punishment against rural families having a
non-permitted second child except a fine. Minority couples were also waved from the
one-child policy in general, even in cities
7
. There were also other variant exemptions in
different provinces, like remarried couples.
Since the early 1990s, the overall fertility rate had decreased below the replacement
level and further declined to 1.22 in the 2000 census and 1.18 in the 2010 census. Due
to concerns about a too-low fertility rate and an aging population, the Chinese govern-
ment cautiously relaxed the one-child policy for limited couples in 2013. This relaxation
allowed all couples in which just one of the husbands or the wife was an only child to have
a second child. After the announcement, all provinces adjusted their local family planning
regulations before mid-2014. This policy change mainly affected urban couples and rural
couples in some provinces in which only one of the husband or wife was an only-child
so they gained permission of having a second child. Due to the limitation of the policy
change, the change in fertility did not meet the government’s forecast. The annual total
new births increased from 16.4 million in 2013 to 16.87 million in 2014, by just 470 thou-
sand, and even declined to 16.55 million in 2015. This led to a broader policy relaxation
soon.
7
As documented by , rural minority couples in all provincial regions except Jiangsu could have a second
child. While urban minority couples not in Fujian, Guangdong, Hunan, Henan, Hubei, Jiangsu, Liaoning,
and Yunnan could have a second child
11
In 2015 the Chinese government announced a universal two-child policy, which per-
mits all couples to have two children, without any restrictions. Moreover, rewards for
single-child families and the pre-birth approval procedure got canceled and the central
government started to encourage different areas to offer subsidies for the second child.
After the universal two-child policy was enacted, the total birth number rose to 17.86 mil-
lion in 2016, a record value since the year 2010. However, the number declined again after
2017 to 17.23 million, 15.23 million, 14.65 million, 12 million, and 10.62 million from 2017
to 2021 respectively. The trajectory of the total number of new birth showed that even
though the relaxation of the one-child policy released some unmet fertility demands, it
can not reverse the declining trends of total new birth. In 2021, the Chinese government
announced a three-child policy, and even though it is still too early to make any statement,
new birth trends show that further relaxation of the birth control policy would have very
little effect on fertility.
1.3 Data Source
The analysis of this paper is primarily based on data from the China Family Panel Survey
(CFPS), a detailed family survey conducted by the Institute of Social Science Survey (ISSS)
of Peking University across 25 provincial areas in China. The national survey started in
2010 and successfully interviewed 14960 families and all family members of these families.
After that, the survey team conducted follow-up surveys every other year to all these in-
dividuals, as well as their children, as “core members”. The current version of the dataset
includes data collected from 2010 to 2020 biannually, and I use the first five-wave data
from 2010 to 2018.
The CFPS survey is a good fit for this paper because, first, it covers 25 provinces
8
. This
makes this dataset both a representative sample of the total population and a convenient
8
Exclude Xinjiang, Xizang, Qinghai, Inner Mongolia, Ningxia, and Hainan.
12
data source since it does not include provinces with special family planning policies. Sec-
ond, the five survey waves I use in this paper are in 2010, 2012, 2014, 2016, and 2018, which
cover two periods before the policy change in 2013, one period between two changes, and
two periods after the universal two-child policy was announced in 2015. Third, this house-
hold survey contains rich demographic and social-economic information so I can use them
to both identify the eligibility of having a second child of each couple and link the eligi-
bility change with their labor market outcomes.
1.3.1 Sample Selection
In my empirical analysis, to capture the impacts of this birth control policy change on fer-
tility and females’ labor supply outcomes, I restrict the samples to only fertile-aged (aged
16 to 45 in 2010) females, who were married in 2010. I include only females married in
2010 because first, the family planning policy regulations are dominantly applicable to
married couples, and to obtain a birth permit, the applicants needed to show a marriage
certificate before 2015, and without a birth permit, a child could not get household reg-
istration (hukou) and had no access to education or social welfare. The other reason for
restricting samples only to married females is that, when I assign treatment status to each
individual, marital status is a determinant factor and I assign each unmarried individual a
treatment status as zero. As a result, an individual can get treated simply by getting mar-
ried before the policy changes, so by restricting to females married in 2010, I can exclude
impacts from females who were just married and focus on the impacts of the one-child
policy relaxation on outcomes of females who were already married in the baseline year.
I also include regression results on males and elder females as robustness checks.
1.3.2 Main Outcomes and Summary Statistics
The outcomes in my empirical part include new birth, working status, weekly work hours,
promotion, and salary earnings. The new birth variables use data from the 2018 wave, and
13
it is defined as an indicator of whether an individual had a new birth in a specific year
or the year before. Thus, the indicator of new birth in the years 2014 and 2016 can also
capture the impact of the policy changes that happened in 2015 and 2013. Working status
is defined by one if an individual is currently working (at least worked one hour in the past
week), or could return to position within 6 months, or in the slack season of agricultural
work or business. Weekly working hours measure the average work time in one week,
including overtime work in the past 12 months in the primary job position. Promotion
is defined to be one if an individual got either an executive promotion or technical title
promotion or both within the past 12 months. Salary earnings are the inverse hyperbolic
sine transformation of the sum of net income from the primary job and all other part-
time jobs, excluding agricultural and business revenues. All these variables measure labor
market outcomes on a yearly level, and my main empirical analysis will also be on a yearly
level.
Table 1.1 shows summary statistics of the main demographic information and labor
market outcomes of females by different treatment groups. The always-treated group fe-
males are those who could have a second child before 2013; the early-treated group females
are those who got permission to have a second child after 2013, and the late-treated group
females are those who were allowed to have a second child only after 2015. This table
shows that, consistent with the regulation details I discuss in the background part, com-
pared with early-treated females, the early-treated females are more likely to live in urban
communities, have an urban hukou, be the only child, have higher education attainments,
have fewer children, and earning more. The late-treated females are more similar to the
early-treated females but still less likely to live in urban areas, have urban hukou, are less
educated, and have more children, and none of them are only-child.
14
Always Treated Early Treated Late Treated
Urban 0.36 0.70 0.49
(0.48) (0.46) (0.50)
Hukou 0.06 0.59 0.30
(0.23) (0.49) (0.46)
Minority 0.18 0.08 0.05
(0.39) (0.27) (0.22)
Only Child 0.07 0.38 0.00
(0.26) (0.49) (0.07)
Education Years 6.20 10.33 7.97
(4.21) (4.45) (4.47)
Child Amount 1.93 1.17 1.45
(0.87) (0.66) (0.77)
Working Status 0.76 0.75 0.74
(0.43) (0.43) (0.44)
Weekly Work Hours 33.31 33.82 33.77
(28.71) (25.69) (28.03)
Promotion 0.02 0.05 0.03
(0.14) (0.21) (0.17)
IHS Salary Earnings 2.99 5.46 4.16
(4.68) (5.39) (5.17)
Observations 14032 1662 19183
Standard errors in parentheses
Table 1.1: Summary Statistics for Fertile-aged Females
Note: All data are from China Family Panel Survey (CFPS), the year 2010 to 2018. The always-treated group contains individuals who
first got treated in 2010 or 2012; the early-treated group contains individuals who first got treated in 2014; the late-treated group
contains individuals who first got treated in 2016 or 2018. I only include females who were between the ages of 16 to 45 and were
married in 2010.
15
1.4 Empirical Strategy
To identify the causal impacts of the relaxation of China’s one-child policy on female la-
bor market outcomes, I employ a generalized Difference in Differences (DID) design with
staggered adoption. This model fits the situation where different groups of individuals
can receive the treatment at different times, and once get treated, will stay treated after-
ward. To show a clear picture of this issue, I will include both a static and a dynamic
specification, and the dynamic model is also the so-called event study method.
1.4.1 Static DID
In a static setup, specifically, I use the following Two-Way Fixed Effects (TWFE) model to
estimate the average treatment effects of whether one is treated in a specific year:
Y
it
=α
i
+θ
t
+D
it
β
treated
+ϵ
it
(1.1)
In this specification, Y
it
is the outcome of the individual i at year t, which could be
either new birth or labor market outcomes; α
i
is individual fixed effects, controlling for
systematic variance in outcomes among different units; θ
t
is year fixed effects, controlling
overall changes in outcomes that are common for all units, like macroeconomic conditions
and parallel national policy changes that affect every individual equally; D
it
is an indicator
of treatment status of individuali at yeart, the eligibility of having a second child with-
out any penalty; β
treated
is the coefficients that I want to estimate here; ϵ
it
is error terms
clustered in individual level.
In the context of this paper, married individuals can be separated into three different
groups, the always-treated group, the early-treated group, and the late-treated group. The
always-treated group, based on the one-child policy regulations before 2013, consists of
all exceptional couples who could have a second child before 2013. I assign a treatment
indicator equal to one from the first period until the last year of the panel to all individuals
16
of this group. The early-treated group consists of couples who were permitted to have a
second child after 2013. The late-treated group consists of all the rest, who got treated in
2015 and stay treated afterward.
1.4.2 Dynamic DID
In the dynamic specification, I employ a generalized TWFE model, which is also called the
event study method, to estimate the average treatment effects of a specific relative year to
the year an individual gets the treatment.
Y
it
=α
i
+θ
t
+
X
r
1[R
it
=r]β
r
+ϵ
it
(1.2)
In specification (2), R
it
is the indicator of relative years to the treatment for individ-
ual i at year t. I normalize the effects at r =−2 to zero, so that all coefficients can be
interpreted as the treatment effects for each value of r relative to the previous survey year
before the treatment. In the context of this paper, the early-treated group individuals have
a sequence of{r} ={−4,−2,0,2,4}, and the late-treated group individuals have a relative
time sequence{r} ={−6,−4,−2,0,2}.
I will mainly show regression results estimating model (1) in the results part and the
heterogeneity analysis part, and include dynamic effects estimates too to show further
term impacts.
1.4.3 Identification Assumptions
To causally identify the treatment effects using either model (1) or model (2), I need two
main assumptions, the no anticipatory effects assumption and the parallel trend assump-
tion (L. Sun and Abraham 2021; Borusyak, Jaravel, and Spiess 2022; Roth, Sant’anna, Bilin-
ski, and Poe 2022).
17
The no anticipatory effects assumption means that before the treatment happens, indi-
viduals were not behaving differently in anticipation of this policy change. In the context
here, it means that neither the early-treated group couples nor the late-treated couples had
an out-of-plan second child facing a penalty or had preparation on the labor market for an
upcoming second child, acting as if they can anticipate a policy change. This is a rational
assumption because, first, the one-child policy in China was rather a serious policy restric-
tion with a large amount of penalty, especially for urban residents, and the conduct of this
policy of local officials was serious too because it was in high priority for local officials’
promotion. The second support comes from the characteristics of policy experimentation
in China summarized by S. Wang and D. Yang 2022. National policy experimentation de-
cisions are dominantly made by the central government and it is unreasonable to believe
couples would systematically anticipate this policy change and started childbearing ear-
lier. In the robustness check part, I use the heterogeneity robust estimation method to test
for joint pre-trends, and the results also show that there were no anticipatory actions.
The parallel trend assumption requires that the outcomes of individuals in different
groups will be on the same trend if there were no treatment status changes. The potential
violation of the parallel trends assumption brings more threats to the identification. A nat-
ural extension of the parallel trend assumption in a simple 2X2 comparison to the parallel
trend assumption in a staggered adoption setting requires that the parallel trend of po-
tential outcomes meet for all 2X2 comparisons between any two groups and any two time
periods. This is a very strong assumption, and, in the context of this paper, can hardly be
satisfied. By definition, during all periods, I have an always-treated group, an early-treated
group, and a late-treated group, separated by their hukou category and province and the
number of siblings of the wife and husband in each couple. Couples and individuals in
these three groups can differ largely in family background and social-economic charac-
teristics, so their potential outcomes could have different trends. In some recent method-
ological literature, researchers have considered variants of the parallel trend assumption.
18
Callaway and Sant’Anna 2021 use a weaker version conditional on covariates; L. Sun and
Abraham 2021 consider a parallel trend assumption using only groups that are eventually
treated, and not groups that never get treated; Borusyak, Jaravel, and Spiess 2022 propose
a stronger version of parallel trends assumption to get a more efficient estimator. In the
robustness check part, I use the estimation method proposed by Callaway and Sant’Anna
2021 and conduct a joint test of all pre-trends. It shows that using that method, which ex-
cludes the always-treated group individuals and only uses not-yet-treated group females
as controls, the parallel trend assumption can be satisfied in a joint test.
1.4.4 Issues On Staggered TWFE Model
As shown in several recent econometric papers, the TWFE estimator I am using, either
static or dynamic, will be biased by treatment heterogeneity among groups and periods (;
Goodman-Bacon 2021; Chaisemartin and D’Haultfœuille 2020; L. Sun and Abraham 2021;
Borusyak, Jaravel, and Spiess 2022; also see reviews by Roth, Sant’anna, Bilinski, and Poe
2022, and Chaisemartin and D’haultfoeuille 2022). This contamination happens because,
in a staggered adoption design model, in both static and dynamic settings, the TWFE es-
timator is a weighted average of all 2X2 comparisons between any pair of groups and
periods. These weights, without any economic interpretation, however, can be negative
because of “Forbidden Comparisons” (Goodman-Bacon 2021). Some of the comparisons
are called “forbidden” because the control group in these comparisons, or the group of in-
dividuals with invariant treatment status between two time periods, can be early treated
groups.
To show that my regression results are not significantly biased by this treatment effects
heterogeneity, I include several practices as robustness checks. First, I will check how seri-
ous is the ”Forbidden comparison” contamination by testing for possible negative weights
19
using the method proposed by Chaisemartin and D’Haultfœuille 2020. Then, I will em-
ploy the new estimation method proposed by Callaway and Sant’Anna 2021, and show
whether regression results change significantly.
1.5 Results
In this section, I will show regression results of the impacts of the one-child policy relax-
ation on both new birth and female labor market outcomes. In the first part of my regres-
sion results, I show that after the fertility policy shock, eligible females are more likely
to have a new birth and less likely to work, work fewer hours each week, and earn less.
The increase in fertility rates is larger among elder age groups, compared with younger
females, and the drop in labor participation is more pronounced among females aged 26
to 35 in the year 2010. Heterogeneous treatment effects analysis shows that
1.5.1 Impacts of Two-child Policy on New Birth
I start the regression part by exploring how effective was this general two-child policy on
the actual new birth. The indicator of the new birth is defined to be one if, in one specific
year, an individual has a child born in this year or the year before. The regression I show
in this part and all the following regressions are restricted to females who were between
the age of 16 to 45 and was married in the year 2010.
In Table 1.2, the first column shows that after the policy shock, 2.5% more eligible fe-
males are having a new birth in a two-year interval. Columns 2 to 4 include regression
results of specifications with subgroups of 16 to 25 years old females in the year 2010, 26
to 35 years old, and also 36 to 45 years old in the year 2010. The increases are larger among
the youngest group, with a 5% increase, and 3.6% and 0.6% increases among the two elder
groups, while the relative increase is larger among the eldest age group, 35 to 45 years old
in the year 2010.
20
(1) (2) (3) (4)
Age 16 to 45 Age 16 to 25 Age 26 to 35 Age 36 to 45
in 2010 in 2010 in 2010 in 2010
Treatment 0.025
∗∗∗
0.049
∗∗
0.036
∗∗∗
0.009
∗∗
(0.006) (0.025) (0.011) (0.004)
Year FE Yes Yes Yes Yes
Individual FE Yes Yes Yes Yes
Observations 31,044 4,632 10,546 14,521
Control Means 0.064 0.271 0.084 0.009
Standard errors in parentheses
∗
p<0.10,
∗∗
p<0.05,
∗∗∗
p<0.01
Table 1.2: Regression Results on New Birth of All Fertile Aged Females and by Three Age
Groups
Note: All data are from China Family Panel Survey (CFPS), the year 2010 to 2018. The outcome variable new birth is defined to be one
if, at this year, the female had a new birth at that year or the year before. Regressions are restricted among females married in 2010,
and results of four groups are shown, aged 16 to 45, 16 to 25, 26 to 35, and 36 to 45 at year 2010. All specifications include individual
fixed effects and year fixed effects. Standard errors are robust and clustered at the individual level.
1.5.2 Impacts of Two-child Policy on Female Labor Market Outcomes
Next, I show the impacts of the one-child policy relaxation on female labor market out-
comes, including labor participation, weekly work hours, promotion, and salary earnings.
Table 1.3 shows that after the fertility policy shock, eligible females are 2.6% less likely to
work, work 1.826 fewer hours per week, and earn 21.1% less annually on salary.
The scale of labor supply decline is very close to the scale of average increase at the new
birth, which shows that the labor supply drop is largely a direct consequence of females
giving more birth after the policy change. However, I will show that besides this direct
consequence, there is also a labor supply drop among females who did not have a new
birth.
Similar to the regression results I showed before of the fertility policy shock’s impact
on the new birth, I also show regression results on labor market outcomes by different
age groups. Table 1.4 shows the regression results of the impacts of the two-child policy
on female labor market outcomes by three age groups, aged 16 to 25, aged 26 to 35, and
21
(1) (2) (3) (4)
Working Status Weekly Work Hours Promotion IHS Salary Earnings
Treatment -0.026
∗∗∗
-1.807
∗∗
-0.004 -0.211
∗∗
(0.009) (0.708) (0.004) (0.093)
Year FE Yes Yes Yes Yes
Individual FE Yes Yes Yes Yes
Observations 30,000 22,986 21,993 29,729
Control Means 0.709 32.046 0.042 3.920
Standard errors in parentheses
∗
p<0.10,
∗∗
p<0.05,
∗∗∗
p<0.01
Table 1.3: Regression Results on Labor Market Outcomes
Note: All data are from China Family Panel Survey (CFPS), the year 2010 to 2018. Regressions are restricted among females married
and among those aged 16 to 45 in the year 2010. Results of regression of four outcome variables are reported, working status, weekly
work hours, promotion, and inverse hyperbolic sine transformation (IHS) of salary earnings. Working status equals one if the
individual worked at least one hour in the past week before the survey time, can return to work within six months, or is during the
business off-season. Weekly work hours are a variable reflecting the total work time in a week of their main job in the past 12 months
at survey time. Promotion is based on a question asking whether the respondent got any promotion at the workplace, either a
technical-level promotion or an executive promotion. IHS earnings measure the total amount of RMB salary the respondent earned in
the past 12 months before the survey time, from both main job and secondary jobs. All specifications include individual fixed effects
and year fixed effects. Standard errors are robust and clustered at the individual level.
aged 36 to 45 in 2010, respectively. There are only significant declines in labor supply and
weekly work hours among the subgroup of females aged 26 to 35 in 2010, even though all
three groups of eligible females have significantly more new birth after the policy shock.
On average, they are 4.2% less likely to work and work 3.509 hours less per week, a 10.2
percent drop in an average of 34.451 hours weekly work hours among this age group. Even
though Table 1.2 shows more increase in new birth happens among the younger group
than the elder group, there is only a slight and insignificant decrease in labor supply and
weekly work hours among this youngest group. If the direct consequence of having a new
birth is the only cause of the female labor supply decline, we should witness the highest
labor participation drop among the youngest group of females, who were 16 to 25 years
old in 2010. However, there is only a significant decline in labor participation among the
elder group of females who were 26 to 35 years old in 2010. This result implies that the
impacts on female labor market outcomes are not only direct consequences of having a
new birth but also a result of household decisions and firm reactions. I will show more
22
heterogeneous impacts in the following parts to explain why the impacts on female labor
supply are different by different groups.
Working Status Weekly Work Hours Promotion IHS Earnings
(1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12)
Treatment -0.019 -0.042
∗∗∗
-0.012 -2.562 -3.509
∗∗∗
-0.366 -0.014 -0.001 -0.005 -0.224 -0.265 -0.142
(0.027) (0.015) (0.011) (1.878) (1.144) (1.008) (0.013) (0.009) (0.005) (0.273) (0.164) (0.124)
Year FE Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes
Individual FE Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes
Observations 4,397 10,147 15,456 3,537 7,804 11,645 3,024 7,424 11,545 4,296 10,080 15,353
Control Means 0.506 0.697 0.753 22.543 31.767 34.034 0.044 0.054 0.036 3.292 4.568 3.684
Standard errors in parentheses
∗
p<0.10,
∗∗
p<0.05,
∗∗∗
p<0.01
Table 1.4: Regression Results on Female Labor Market Outcomes by Age Groups
Note: All data are from China Family Panel Survey (CFPS), the year 2010 to 2018. Regressions are restricted among females married
and among those aged 16 to 45 in the year 2010. Results of regression of four outcome variables are reported, working status, weekly
work hours, promotion, and inverse hyperbolic sine transformation (IHS) of salary earnings. Working status equals one if the
individual worked at least one hour in the past week before the survey time, can return to work within six months, or is during the
business off-season. Weekly work hours are a variable reflecting the total work time in a week of their main job in the past 12 months
at survey time. Promotion is based on a question asking whether the respondent got any promotion at the workplace, either a
technical-level promotion or an executive promotion. IHS earnings measure the total amount of RMB salary the respondent earned in
the past 12 months before the survey time, from both main job and secondary jobs. All specifications include individual fixed effects
and year fixed effects. Standard errors are robust and clustered at the individual level.
1.5.3 Heterogeneous Impacts of Two-child Policy by Child Amount
In this subsection, I will show the results of heterogeneity analysis by the number of chil-
dren in 2010. The lift of the strict one-child policy has the most direct influence on families
with one child already but not allowed to have a second one under the restriction. Hence,
one should expect to see larger impacts of this policy shock on females who had one child
in the baseline year.
Table 1.5 shows the heterogeneous impacts of the two-child policy shock on female
fertility and labor market outcomes, by the number of children they had before the policy
changed. Column (1) shows that after the policy change, eligible females who had two
children or more in 2010 are still more likely to have a new birth. Even though the general
two-child policy did not mark the termination of the family planning regime in China, the
policy change in 2015 largely relaxed population control, canceled subsidies to single-child
parents and canceled pre-birth approval processes. Some provinces started to provide
23
(1) (2) (3) (4) (5)
New Birth Working Status Weekly Work Hours Promotion IHS Earnings
Treatment 0.042
∗∗∗
-0.012 -0.453 0.007 -0.280
∗∗
(0.005) (0.011) (0.964) (0.004) (0.123)
Treatment× Having -0.100
∗∗∗
0.023 2.138 -0.016 0.109
No Child (0.020) (0.025) (1.776) (0.013) (0.267)
Treatment× Having -0.012
∗
-0.033
∗∗
-3.178
∗∗∗
-0.019
∗∗∗
0.114
One Child (0.007) (0.013) (1.111) (0.006) (0.144)
Year FE Yes Yes Yes Yes Yes
Individual FE Yes Yes Yes Yes Yes
Observations 31,044 30,000 22,986 21,993 29,729
Control Means 0.064 0.709 32.046 0.042 3.920
Standard errors in parentheses
∗
p<0.10,
∗∗
p<0.05,
∗∗∗
p<0.01
Table 1.5: Heterogeneous Impacts by Child Amount in 2010
Note: All data are from China Family Panel Survey (CFPS), the year 2010 to 2018. Regressions are restricted among females married
and among those aged 16 to 45 in the year 2010. Results of regression of four outcome variables are reported, working status, weekly
work hours, promotion, and inverse hyperbolic sine transformation (IHS) of salary earnings. Working status equals one if the
individual worked at least one hour in the past week before the survey time, can return to work within six months, or is during the
business off-season. Weekly work hours are a variable reflecting the total work time in a week of their main job in the past 12 months
at survey time. Promotion is based on a question asking whether the respondent got any promotion at the workplace, either a
technical-level promotion or an executive promotion. IHS earnings measure the total amount of RMB salary the respondent earned in
the past 12 months before the survey time, from both main job and secondary jobs. All specifications include individual fixed effects
and year fixed effects. Standard errors are robust and clustered at the individual level.
with fertility encouraging benefits , and all of these acts contributed to the increase in new
birth among families who already had more than one child before the policy changed.
There are also significant increases in new birth among females who had just one child in
2010, but relatively less than the first groups. The rising birth rate among one-child policy
compliers suggests that policy effect was part of the reason why they chose to have only
one child before, and the lift of fertility restrictions released some unmet child demand.
On the other hand, females with no child in 2010 are less likely to have a new birth after the
policy change. This group of married females is relatively younger, and better educated
than the other two groups. After the policy changes, they are more willing to delay the
birth of their first child and shorten the interval between the first two children. The one-
child policy lift could affect the life circle decision of young females of education, family,
fertility, and career, and this can help explain the short-term decline in the birth rate among
them.
24
Columns (2) to (4) of Table 1.5 show that compared with females with more than one
child in 2010, females who had just one child in 2010 experienced a significantly larger
drop in labor participation (4.5 percent points decline), weekly work hours (a 3.639 hours
decline) and promotion (1.2 percentage points drop). There is a significant decline in
salary earnings among females who had more than one child in 2010, but no significant
impacts on other labor market outcomes. The finding that the one-child policy complying
females having fewer new birth but got affected more on their labor supply shows that the
consequences of having a new birth or just higher fertility expectations can be varied by
different price effects or income effects. Higher opportunity cost can help to explain both
the decision of having just one child under the one-child constraint and the bigger impact
of increasing fertility demand and expectation on labor participation. The higher income
effect also makes them more likely to leave their work and focus more on child care. I will
explain these two effects in detail in the mechanism part.
1.5.4 Heterogeneous Impacts of Two-child Policy by Gender of First
Child
In this section, I present the results of a regression analysis exploring the heterogeneous
treatment effects of the one-child policy relaxation based on child gender among females
who complied with the policy. It is expected that due to son preferences, women with only
one daughter would be more willing to take advantage of the policy relaxation, leading to
a larger decrease in their future labor supply as their fertility expectations increase.
Table 1.6 demonstrates that, for women with a single child in 2010, eligible mothers
with only a son are less likely to give birth to a new child following the policy change, al-
though the difference is not statistically significant. These mothers exhibit a 61% smaller
decline in labor supply (-0.039 and -0.100), as well as a reduction in weekly work hours
that is approximately half as large (-3.846 hours and -6.892 hours). The varied responses
in labor supply indicate that, while both women with one daughter and one son are more
25
likely to have a new child after the policy change, with similar increases in actual fertility,
their labor supply and work hours are more significantly impacted by the policy change.
This evidence suggests that the decrease in female labor supply not only stems from in-
creased fertility and maternity leave but also potentially reflects differing treatment for
mothers with an only son compared to those with an only daughter in the labor market.
(1) (2) (3) (4) (5)
New Birth Working Status Weekly Work Hours Promotion IHS Earnings
Treatment 0.068
∗∗∗
-0.100
∗∗∗
-6.892
∗∗∗
-0.025
∗
-0.060
(0.013) (0.019) (1.588) (0.014) (0.224)
Treatment× First Child -0.009 0.061
∗∗∗
3.046
∗
0.029
∗∗
-0.317
is a Son (0.012) (0.019) (1.641) (0.014) (0.227)
Test :β
1
+β
2
= 0 0.000 0.008 0.002 0.617 0.021
Year FE Yes Yes Yes Yes Yes
Individual FE Yes Yes Yes Yes Yes
Observations 13,932 13,522 10,543 9,958 13,380
Control Means 0.061 0.721 33.972 0.056 4.952
Standard errors in parentheses
∗
p<0.10,
∗∗
p<0.05,
∗∗∗
p<0.01
Table 1.6: Heterogeneous Impacts by Gender of the First Child in 2010
Note: All data are from China Family Panel Survey (CFPS), the year 2010 to 2018. Regressions are restricted among females married
and among those aged 16 to 45 in the year 2010, and also to those who already and only had one child. Results of regression of four
outcome variables are reported, working status, weekly work hours, promotion, and inverse hyperbolic sine transformation (IHS) of
salary earnings. Working status equals one if the individual worked at least one hour in the past week before the survey time, can
return to work within six months, or is during the business off-season. Weekly work hours are a variable reflecting the total work time
in a week of their main job in the past 12 months at survey time. Promotion is based on a question asking whether the respondent got
any promotion at the workplace, either a technical-level promotion or an executive promotion. IHS earnings measure the total amount
of RMB salary the respondent earned in the past 12 months before the survey time, from both main job and secondary jobs. All
specifications include individual fixed effects and year fixed effects. Standard errors are robust and clustered at the individual level.
1.5.5 Robustness Check
I conduct a series of checks to show my results are robust to alternative specifications, and
they are not caused by potential confounders.
1.5.5.1 Results Using Alternative Estimation Methods
To echo the recent concern about potential bias caused by treatment effects heterogene-
ity while estimating the average treatment effects in a difference in differences design,
26
especially in a staggered adoption design, I also conduct heterogeneity robust methods
proposed by . Their method uses only not-yet-treated or never treated individuals as con-
trols, and it can get regression results with less precision, but free of bias due to treatment
heterogeneity either in time or groups.
First, to test how serious is the potential negative weights problem in my design, I use
the method proposed by Chaisemartin and D’Haultfœuille 2020
9
to calculate the propor-
tion of comparisons with negative weights and the scale of the sum of negative weights.
Test results show that in regression on four main outcomes, including new birth, working
status, weekly work hours, and IHS earnings, the average treatment effect in each model
is a weighted sum of 15 average treatment effects on treated (ATT) of one specific group
and year pair of comparison, while 6 of them receive negative weight
10
. The sums of the
negative weights are -0.836, -0.836, -0.807, and -0.832. In the regression on promotion, 4
out of 13 ATTs receive negative weight, and the sum is -0.861. Because all weights should
sum to one, these test results indicate that the negative weights could be problematic.
I tested the robustness of my results using an alternative estimator, the “group-time av-
erage treatment effects estimator” proposed by Callaway and Sant’Anna 2021
11
. The main
intuition is that the estimator uses observations never-treated or not-yet-treated treated as
controls, so it can avoid the bias caused by using already treated groups as controls. In the
context of this paper, after 2015, the general two-child policy allows everyone to have two
children, so there are no never-treated observations. Hence, observations in group 3, who
were treated in 2015, are the only not-yet-treated before 2016, and there are no controls
available to evaluate the second policy change in 2015 using this method. This method
also ignores the always-treated observations and excludes all years after the selected con-
trol group received the treatment, so it will only use the first policy change to estimate
9
Using Stata code “twowayfeweights” provided by
10
By design, there should be only three periods when an individual first gets treated, 2010, 2014 and 2018,
however, some individual’s first period of being treated is 2012 or 2016. This happens because of missing
values in specific years.
11
Using Stata code “csdid” provided by
27
the average treatment effects. Estimators using this method can avoid bias from treatment
effects heterogeneity among different groups treated at different times, but could also lose
precision since sample selection.
Results using the “group-time average treatment effects estimator” method show that
the average treatment effect on the treated is 0.038 (with a standard error of 0.065) on
the new birth, -0.0016 (with a standard error of 0.091) on working status, -10.972 (with
a standard error 6.136) on weekly work hours, and -0.622 (with a standard error 0.786).
All of these coefficients estimated using the method proposed by is still in the same sign
as the estimators using the conventional TWFE model, but less precise except for the co-
efficient on treatment effect on weekly work hours. This exercise shows that treatment
heterogeneity across different groups treated at different times does exist in the context,
but it is unlikely to greatly bias my results.
1.5.5.2 Other Robustness checks
I also conducted some other practices to show my main findings are robust.
I show regression results of the treatment effects on elder females and males. If the
changes in new birth and female labor supply are due to responses to the lift of the one-
child policy, I shall expect to see no different impacts on elder females who were above age
45 in 2010 after they were capable to have a second child. Moreover, if the drops in labor
supply and earnings are the results of the household’s response to the lift of the one-child
policy, then based on previous empirical findings on unequal parental responsibilities, I
shall see the same impacts on new birth and smaller impacts on labor supply and earnings
of husbands. Table 1.7 shows the treatment effects on fertile-aged males and elder females,
who were between 45 and 60 in 2010. It indicates that married males at fertile age (16 to
45 in 2010) are also having more new birth, which is a natural result, however, the decline
in working status is smaller, with no significant drop in weekly work hours, and earnings.
There are still some impacts on the rate of getting a promotion, and on earnings, but the
28
drop in labor supply is smaller. This is evidence of differential bargaining power within a
household and unequal parenthood responsibilities. Results using the other specification
in table 1.7 also show that the policy change has no impact on elder females’ (above 45 but
below 60 in 2010) labor participation and work hours.
New Birth Working Status Weekly Work Hours Promotion IHS Earnings
(1) (2) (3) (4) (5) (6) (7) (8) (9) (10)
Treatment 0.022
∗∗∗
0.000 -0.018
∗∗
-0.008 -0.236 -0.182 -0.017
∗∗∗
-0.006
∗∗∗
-0.064 -0.248
∗∗∗
(0.006) (0.000) (0.007) (0.011) (0.709) (0.707) (0.006) (0.002) (0.110) (0.078)
Year FE Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes
Individual FE Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes
Males Yes Yes Yes Yes Yes
Age above 45 in 2010 Yes Yes Yes Yes Yes
Observations 26,473 21,180 25,610 20,455 18,785 16,495 18,361 15,460 25,330 20,383
Control Means 0.069 0.001 0.850 0.554 42.657 21.721 0.077 0.016 5.984 1.595
Standard errors in parentheses
∗
p<0.10,
∗∗
p<0.05,
∗∗∗
p<0.01
Table 1.7: Treatment Effects on Males and Elder Females
Note: All data are from China Family Panel Survey (CFPS), the year 2010 to 2018. In each outcome group, the first specification shows
results on males and the second one on females aged above 45 but below 60 in 2010. Results of regression of four outcome variables
are reported, working status, weekly work hours, promotion, and inverse hyperbolic sine transformation (IHS) of salary earnings.
Working status equals one if the individual worked at least one hour in the past week before the survey time, can return to work
within six months, or is during the business off-season. Weekly work hours are a variable reflecting the total work time in a week of
their main job in the past 12 months at survey time. Promotion is based on a question asking whether the respondent got any
promotion at the workplace, either a technical-level promotion or an executive promotion. IHS earnings measure the total amount of
RMB salary the respondent earned in the past 12 months before the survey time, from both main job and secondary jobs. All
specifications include individual fixed effects and year fixed effects. Standard errors are robust and clustered at the individual level.
1.5.6 Dynamic Effects and Parallel Trends
In this section, I present the dynamic effects of the one-child policy relaxation on female
fertility and labor market outcomes using Model (2), as well as results from a parallel
trends check using the same model. Due to potential negative-weight concerns identified
in the robustness check, I employ the heterogeneity robust estimation method proposed
by (Callaway and Sant’Anna 2021) in the context of Model (2)
12
. This method calculates
the sum of all group-specific effects at different relative years and displays the results in
an event study graph
13
.
12
Using Stata code “csdid” provided by (Callaway and Sant’Anna 2021)
13
Using Stata code “event plot” provided by Borusyak, Jaravel, and Spiess 2022
29
The method suggested by Callaway and Sant’Anna 2021 employs only not-yet-treated
groups as controls, disregarding always-treated groups in the regression. After 2016, all
groups were treated, leaving no not-yet-treated individuals, causing the method unable
to estimate the specific treatment effects of the 2015 policy change. For the 2013 policy
change, the method utilizes late-treated groups as controls, enabling the estimation of
all available dynamic effects. Figures 1.1 display the dynamic effects estimated for new
births, working status, weekly work hours, and IHS salary earnings, using specifications
from Model (2). Long-term effects, 6 years post-treatment, are estimated imprecisely due
to the inherent lack of information available to estimate dynamic effects after 6 years using
not-yet-treated groups as controls. Estimates in this case can be disregarded.
The event study graph in (a) of 1.1 reveals a slightly positive effect on the probability of
having a new birth in the same year a female is treated, with the impact becoming negative
two years later. This indicates that the increase in new births primarily occurs immediately
following the relaxation announcement, suggesting a substantial amount of unmet fertility
demand. However, there is no further increase in new births beyond the first 2 years. The
dynamic estimates figure for working status demonstrates a related trend, with the drop in
labor supply mainly occurring two years after the policy change. Subfigures (c) and (d) in
1.1 display similar dynamic effect patterns, with (c) illustrating a slight reduction in work
time upon policy change announcement and no impacts thereafter, and (d) showing a
minor decline in salary earnings with no further impacts afterward.
Due to the limited sample size when using the heterogeneity robust estimation method,
all estimates are imprecise and do not provide meaningful dynamic average treatment
effects results. Nevertheless, the event study figures offer suggestive evidence that the
increase in fertility following the one-child policy relaxation was limited and primarily
occurred immediately after the policy change, with the early responses in labor supply
and earnings potentially contributing to the restricted encouragement of new births.
30
Utilizing the same event study graphs, I can also demonstrate that pre-trends of treated
and control groups (not-yet-treated) are parallel. The figures 1.1 display dynamic im-
pacts of treatment status on fertility and labor market outcomes before and after the policy
shock. All four subfigures provide no evidence of pre-treatment differences in new births,
labor supply, work hours, or earnings. As the dynamic estimation in 1.1 using the method
by only considers the first policy shock in 2013, these results indicate no evidence of a
violation in the parallel trends assumption, at least for the first stage of policy shock in
2013.
(a) New Birth (b) Working Status
(c) Weekly Work Hours (d) IHS Earnings
Figure 1.1: Dynamic Impacts
Note: All data are from China Family Panel Survey (CFPS), the year 2010 to 2018. Regressions are restricted among females married
and aged 16 to 45 in 2010. All specifications include individual fixed effects and year fixed effects. Standard errors are robust and
clustered at the individual level. These figures show dynamic estimates using the Stata package “csdid” proposed by , and the figures
were drawn using Stata package “event plot” proposed by Borusyak, Jaravel, and Spiess 2022.
31
1.6 Mechanisms of Heterogeneous Effects on One-child Policy
Compliers
There are a number of potential mechanisms behind the previous finding that the impacts
of the fertility policy relaxation on one-child policy compliers are more pronounced, and
I can use the information available to test some of them empirically. Females who had just
one child in 2010 were more educated and were more likely to do non-agricultural work,
so the potential value of the lost labor supply before and after childbirth is higher for
them. Besides this price effect, higher family economic support, and lower family support
in child caring can also lead to the decision of more labor supply drop. One-child policy
compliers are more likely to live in urban communities and own an urban hukou, and they
are from families with smaller sizes, so less support from families in child care is available.
I will use the following triple-difference model to estimate differential response to policy
shock by education attainments, occupation, urban or rural, and family average income:
Y
it
=α
i
+θ
t
+β
1
D
it
+β
2
D
it
×ChildAmount
it
+β
3
D
it
X
it
+β
4
D
it
×ChildAmount
it
×X
it
+β
5
ChildAmount
it
×X
it
+β
6
ChildAmount
it
+β
7
X
it
+ϵ
it
(1.3)
In this model, X
it
are individual’s information that could affect her response to the
relaxation of the one-child policy, including education attainment in 2010, whether living
in an urban community or not in 2010, whether having an urban hukou or not in 2010,
whether doing the non-agricultural job or not in 2010, and whether from families with
average income above the median in 2010. To avoid the endogenous choice of the above
characteristics, I only use that information in 2010, when it was before any relaxation of the
one-child policy. As a result,β
5
,β
6
,andβ
7
will all be included in individual fixed effects,
and estimates ofβ
4
can indicate differential impacts within the groups of females with one
child or no child by different attributes.
32
(1) (2) (3) (4) (5)
New Birth Working Status Weekly Work Hours Promotion IHS Earnings
Treatment 0.038
∗∗∗
0.006 0.707 0.006 -0.339
∗∗
(0.006) (0.013) (1.175) (0.004) (0.140)
Treatment× Having No Child -0.047 -0.063 0.367 -0.027 -0.415
× Junior High or Higher (0.052) (0.065) (4.094) (0.023) (0.662)
Treatment× Having One Child 0.030
∗∗
-0.032 -2.671 -0.032
∗∗∗
-0.368
× Junior High or Higher (0.013) (0.028) (2.466) (0.012) (0.324)
Year FE Yes Yes Yes Yes Yes
Individual FE Yes Yes Yes Yes Yes
Observations 25,960 25,384 19,681 19,184 25,286
Control Means 0.064 0.709 32.046 0.042 3.920
Standard errors in parentheses
∗
p<0.10,
∗∗
p<0.05,
∗∗∗
p<0.01
Table 1.8: Heterogeneous Impacts by Education Attainment in 2010
Note: All data are from China Family Panel Survey (CFPS), the year 2010 to 2018. Regressions are restricted among females married
and among those aged 16 to 45 in the year 2010. Results of regression of four outcome variables are reported, working status, weekly
work hours, promotion, and inverse hyperbolic sine transformation (IHS) of salary earnings. Working status equals one if the
individual worked at least one hour in the past week before the survey time, can return to work within six months, or is during the
business off-season. Weekly work hours are a variable reflecting the total work time in a week of their main job in the past 12 months
at survey time. Promotion is based on a question asking whether the respondent got any promotion at the workplace, either a
technical-level promotion or an executive promotion. IHS earnings measure the total amount of RMB salary the respondent earned in
the past 12 months before the survey time, from both main job and secondary jobs. Variables no child or one child was based on an
individual’s child amount in 2010, and the variable junior or higher was defined to be one if an individual finished at least junior
school by 2010. All specifications include individual fixed effects and year fixed effects. Standard errors are robust and clustered at
the individual level.
33
I test the price effect of fertility decisions and how it affects the differential response
to the policy change by education attainment using model (3). I separate individuals by
whether they had finished junior high school or not by 2010, which corresponds to 8 to
9 years of education. Column (1) of table 1.8 shows that within one-child policy com-
pliers, better-educated females have 3 percentage points more new birth in a two-year
interval, compared with their lower-educated cohorts. Columns (2) and (3) of table 1.8
show that the responses in labor supply to policy change among better-educated one-child
policy compliers are also slightly higher, and column (4) shows that they also received
significantly fewer promotions, compared with lower-educated cohorts. One-child pol-
icy compliers received more education than females with more than one child, and the
larger responses to policy change on new birth and labor supply are concentrated among
the better-educated subgroup, according to results in table 1.8. Better-educated females
have higher earnings, which means a higher opportunity cost a drop in labor supply, and
a higher price effect in the decision of childbirth. Higher educated females being more
affected by the policy change can also be a result of their higher willingness to spend time
on child-caring, thus more labor supply decline.
Table 1.9 shows the results of heterogeneous impacts by the community category an
individual lives in or her hukou category in 2010. I still use the community category in
the baseline survey year, which was before the first policy was changed in 2013, to avoid
potential bias caused by migration. Moreover, since 2009, the National Bureau of Statistics
(NBS) of China publishes community IDs with urban or rural division codes annually,
and some communities were reclassified from rural to urban each year (Gan, Q. He, Si,
and D. Yi 2019)
14
. I can also avoid potential bias caused by community reclassification
using just information in 2010. Similarly, the hukou category was the main determin-
ing factor of eligibility for having a second child under the one-child policy restriction, I
14
claims that some of the communities reclassified from rural to urban just simply due to actual construc-
tion, which changes attribute of land contiguity, rather than economic development. As a result, they show
that 33.4% of total urban population growth from 2010 to 2015 comes from community reclassification.
34
New Birth Working Status Weekly Work Hours Promotion IHS Earnings
(1) (2) (3) (4) (5) (6) (7) (8) (9) (10)
Treatment 0.044
∗∗∗
0.042
∗∗∗
-0.011 -0.008 -1.626 -0.691 0.009
∗∗
0.008
∗
-0.373
∗∗∗
-0.286
∗∗
(0.006) (0.006) (0.012) (0.012) (1.085) (0.995) (0.004) (0.004) (0.137) (0.128)
Treatment× Having No Child 0.006 -0.086
∗
-10.279
∗∗∗
-0.071
∗∗∗
-0.496
× Living in Urban Area (0.041) (0.052) (3.661) (0.027) (0.562)
Treatment× Having One Child 0.048
∗∗∗
-0.085
∗∗∗
-9.985
∗∗∗
-0.013 -0.531
∗
× Living in Urban Area (0.014) (0.029) (2.404) (0.013) (0.313)
Treatment× Having No Child 0.011 -0.083 -13.022
∗∗∗
-0.078
∗∗
-0.769
× Urban Hukou (0.043) (0.063) (4.460) (0.032) (0.665)
Treatment× Having One Child 0.036
∗∗
-0.063 -9.236
∗∗∗
-0.013 -0.231
× Urban Hukou (0.014) (0.039) (3.132) (0.017) (0.396)
Year FE Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes
Individual FE Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes
Observations 30,925 30,852 29,898 29,857 22,913 22,876 21,923 21,897 29,633 29,585
Control Means 0.064 0.064 0.709 0.709 32.046 32.046 0.042 0.042 3.920 3.920
Standard errors in parentheses
∗
p<0.10,
∗∗
p<0.05,
∗∗∗
p<0.01
Table 1.9: Heterogeneous Impacts by Community or Hukou Category in 2010
Note: All data are from China Family Panel Survey (CFPS), the year 2010 to 2018. Regressions are restricted among females married
and among those aged 16 to 45 in the year 2010. Results of regression of four outcome variables are reported, working status, weekly
work hours, promotion, and inverse hyperbolic sine transformation (IHS) of salary earnings. Working status equals one if the
individual worked at least one hour in the past week before the survey time, can return to work within six months, or is during the
business off-season. Weekly work hours are a variable reflecting the total work time in a week of their main job in the past 12 months
at survey time. Promotion is based on a question asking whether the respondent got any promotion at the workplace, either a
technical-level promotion or an executive promotion. IHS earnings measure the total amount of RMB salary the respondent earned in
the past 12 months before the survey time, from both main job and secondary jobs. Variables of no child or one child were based on
each individual’s child amount in 2010. Urban is defined to be one if an individual lives in a community categorized as urban in 2010,
and hukou was defined to be one if an individual owns a non-agricultural hukou in 2010. All specifications include individual fixed
effects and year fixed effects. Standard errors are robust and clustered at the individual level.
can also avoid potential bias due to the self-chosen change of the hukou category. I also
include regression results excluding individuals who changed their hukou category or
community category in the robustness check part. Table 1.9 shows that among one-child
policy compliers, females who were living in urban communities have an average birth
rate 4.8 percentage points higher than those who were living in rural communities, after
the policy change for them. Meanwhile, they are also more likely to leave their work (8.5
percentage points more), work fewer hours per week (9.973 hours fewer per week), and
earn less salary income, compared with one-child policy compliers living in rural com-
munities. Similar results among one-child policy compliers who had a non-agricultural
hukou were found too, with a smaller magnitude. These results suggest that compared
with rural females who chose to obey the one-child policy restriction, urban one-child pol-
icy compliers respond more to the policy relaxation and experienced more drop in labor
supply and work hours. This can be a result of both higher price effect and higher income
35
effect, due to higher income loss and higher family economic support, but less support in
child-caring due to smaller family size.
(1) (2) (3) (4) (5)
New Birth Working Status Weekly Work Hours Promotion IHS Earnings
Treatment 0.044
∗∗∗
-0.128
∗∗∗
-5.380
∗∗∗
0.016
∗∗∗
-0.215
(0.006) (0.013) (1.337) (0.004) (0.175)
Treatment× Having No Child -0.078 0.108 0.289 -0.023 -0.935
× Doing Non-agricultural Work (0.067) (0.076) (5.490) (0.018) (0.799)
Treatment× Having One Child 0.000 -0.082
∗∗∗
-1.710 -0.024
∗∗
-0.119
× Doing Non-agricultural Work (0.015) (0.026) (2.562) (0.010) (0.334)
Year FE Yes Yes Yes Yes Yes
Individual FE Yes Yes Yes Yes Yes
Observations 25,936 25,360 19,660 19,166 25,262
Control Means 0.064 0.709 32.046 0.042 3.920
Standard errors in parentheses
∗
p<0.10,
∗∗
p<0.05,
∗∗∗
p<0.01
Table 1.10: Heterogeneous Impacts by Occupation in 2010
Note: All data are from China Family Panel Survey (CFPS), the year 2010 to 2018. Regressions are restricted among females married
and among those aged 16 to 45 in the year 2010. Results of regression of four outcome variables are reported, working status, weekly
work hours, promotion, and inverse hyperbolic sine transformation (IHS) of salary earnings. Working status equals one if the
individual worked at least one hour in the past week before the survey time, can return to work within six months, or is during the
business off-season. Weekly work hours are a variable reflecting the total work time in a week of their main job in the past 12 months
at survey time. Promotion is based on a question asking whether the respondent got any promotion at the workplace, either a
technical-level promotion or an executive promotion. IHS earnings measure the total amount of RMB salary the respondent earned in
the past 12 months before the survey time, from both main job and secondary jobs. Variables of no child or one child were based on
each individual’s child amount in 2010. Non-agricultural was defined to be one if an individual’s main job was not farming on her
own land, including employed farming, in 2010. All specifications include individual fixed effects and year fixed effects. Standard
errors are robust and clustered at the individual level.
Table 1.10 shows a similar mechanism, that among one-child policy compliers, females
whose main job was not farming on their own land in 2010 do not have a more new birth,
but are significantly less likely to work (8.2% less), work slightly fewer hours per week,
and less likely to get promoted. These results imply that compared with females who had
just one child under the one-child policy and were doing agricultural work in 2010, the
negative impacts of the policy relaxation on labor supply only show up among those who
were employed or self-employed. Individuals doing non-agricultural work have higher
opportunity costs to have one more child, because of less job flexibility and high income.
Females from families with more income should have higher economic support for
child expenses and also living expenses with income lost due to childbirth. Table 1.11
shows that females in families with a higher average income in 2010 are 4.2% more likely
36
to react to the fertility policy relaxation and have another child in a two-year interval, after
the policy shock, and also less likely to work (5% less), work fewer hours (insignificant
3.274 hours fewer). The differential responses to the fertility policy change among one-
child policy compliers by family income provide a piece of evidence that different income
effects in the fertility decision lead to different reactions to the fertility restriction lift.
(1) (2) (3) (4) (5)
New Birth Working Status Weekly Work Hours Promotion IHS Earnings
Treatment 0.036
∗∗∗
0.008 1.297 0.009
∗
-0.143
(0.006) (0.013) (1.148) (0.005) (0.146)
Treatment× Having No Child 0.014 -0.124
∗∗
-6.926
∗
-0.068
∗∗
-1.580
∗∗∗
× Family Income Higher than Median (0.042) (0.051) (3.652) (0.027) (0.532)
Treatment× Having One Child 0.042
∗∗∗
-0.050
∗
-3.274 -0.015 -0.487
× Family Income Higher than Median (0.014) (0.028) (2.399) (0.012) (0.311)
Year FE Yes Yes Yes Yes Yes
Individual FE Yes Yes Yes Yes Yes
Observations 29,399 28,429 21,804 20,901 28,169
Control Means 0.064 0.709 32.046 0.042 3.920
Standard errors in parentheses
∗
p<0.10,
∗∗
p<0.05,
∗∗∗
p<0.01
Table 1.11: Heterogeneous Impacts by Average Family Income in 2010
Note: All data are from China Family Panel Survey (CFPS), the year 2010 to 2018. Regressions are restricted among females married
and among those aged 16 to 45 in the year 2010. Results of regression of four outcome variables are reported, working status, weekly
work hours, promotion, and inverse hyperbolic sine transformation (IHS) of salary earnings. Working status equals one if the
individual worked at least one hour in the past week before the survey time, can return to work within six months, or is during the
business off-season. Weekly work hours are a variable reflecting the total work time in a week of their main job in the past 12 months
at survey time. Promotion is based on a question asking whether the respondent got any promotion at the workplace, either a
technical-level promotion or an executive promotion. IHS earnings measure the total amount of RMB salary the respondent earned in
the past 12 months before the survey time, from both main job and secondary jobs. Variables of no child or one child were based on
each individual’s child amount in 2010. Variable family income higher than the median was defined to be one if an individual’s
family income per person is higher or equal to the median level in 2010. All specifications include individual fixed effects and year
fixed effects. Standard errors are robust and clustered at the individual level.
Those four heterogeneity analyses within one-child policy compliers using the triple-
difference method show that being better educated, living in an urban community or
owning urban hukou, doing non-agricultural jobs, and coming from families with higher
average income are all correlated with less response to the one-child policy relaxation
and higher potential child penalty on labor market outcomes. Better educational attain-
ment and non-agricultural work are both correlated with higher wage loss and living in
cities is correlated with smaller family sizes. These all lead to higher opportunity costs of
having one more child, with more earning loss and less family support for child-caring.
37
Higher family average income means that the family can provide more economic support
for child-raring and can afford income loss after childbirth. After the lift of the one-child
policy restriction, both price effects and income effects can affect a household’s fertility
decision as a response to it and further impact female labor supply and earnings. This
can also help explain why the increase in fertility was lower than the government’s fore-
cast and as the new-generation females are higher educated, and more concentrated in
cities, the intention to further increase the overall fertility rate by further relaxing policy
restrictions can hardly be achieved.
1.7 Conclusion
After more than 35 years, China changed its famous and highly controversial one-child
policy to a two-child policy in 2015, with the clear intention to reverse the long-term trend
of low fertility rate and population aging. The responses of different families to this policy
change, however, are still unclear, as well as the potential further impacts on female labor
market outcomes.
In this paper, I use a family panel survey, the CFPS survey dataset, and exploit detailed
policy regulations in each province covered in the survey and combine them with demo-
graphic and social-economic information. Using mainly hukou category and province,
ethnicity, sibling number, and marital status information, I assign a treatment status of
eligibility for having a second child to all couples with available information within the
survey, from 2010 to 2018. Focusing on the policy change in both 2013 and 2015, I can use
timing variation on an individual level to causally identify responses to the one-child pol-
icy relaxation among different couples, and the impacts of female labor supply and other
labor market outcomes.
Regression results show that on average, eligible females are having more new birth,
and are also less likely to work, work fewer hours, and earn less. The increase in the
38
new birth is more pronounced among elder cohorts, and the drop in labor supply only
shows up among prime-aged females (aged 26 to 35 in 2010). Moreover, compared with
females with no child or more than one child before the policy changed, the one-child
policy complying females with just one child is less likely to respond to the policy change
by having one more child. However, they are the only group of females who experienced
a significant drop in labor supply, work hours, and promotion. The differential responses
among one-child policy compliers and larger impacts on labor market outcomes can both
be explained by higher opportunity costs of having a second child, more family economic
support, and less family support on child-caring. Both the price effects and income ef-
fects are contributing to the smaller fertility demand and higher expected child penalty
due to their higher education attainment, higher probabilities of living in cities, owning
urban hukou, doing non-agricultural work, and coming from families with higher average
income.
All results are robust either using the new heterogeneity robust estimation method
or using different robustness checks. Dynamic treatment effects analysis also shows that
the drop in female labor supply and work time does not persist after the first two to four
years, mainly because of lower realized fertility increase from one-child policy complying
couples. The findings in this paper provide a comprehensive analysis of the individual re-
sponse to a more generous fertility policy and further impacts on labor market outcomes,
especially for females. This paper brings insights into the idea that government popula-
tion policies should be collaborative with female employee protection and support. In the
Chinese context, this paper implies that the sole policy change from the universal two-
child policy to a three-child policy in 2021 can hardly reverse the downward fertility rate
trend, but may bring unbalanced incentives for childbearing among lower-tier females.
39
Chapter 2
Income Inequality and Heterogeneous Return to Education
in China
1
2.1 Introduction
Following the economic reforms in 1978, China witnessed remarkable economic growth
and has now become the second-largest economy in the world. However, alongside this
soaring economy, income inequality has also risen significantly, raising continuous con-
cerns from researchers worldwide and the Chinese government itself (Ravallion and Shao-
hua Chen 2007, Piketty and Qian 2009, Y. Xie and X. Zhou 2014, Knight 2014, Piketty, L.
Yang, and Zucman 2019, J. Zhang 2021, as well as the State Council, 2013
2
). Utilizing es-
timates from various sources, China’s Gini coefficient has increased from approximately
0.30 in 1978 to 0.47 in 2016 (J. Zhang 2021).
1
Co-authored with Zhan Gao
2
In 2013, the State Council announced ”the opinions on deepening the reform of the income distribution
system,” which recognized the urgent problem of rural-urban gaps, regional gaps, and income inequality
and announced a comprehensive set of policies to achieve income equity. http://www.gov.cn/zwgk/2013
-02/05/content_2327531.htm (in Chinese).
40
Figure 2.1: GDP per capita and Gini Index in China Since 1978
Note: GDP per capita data is from the World Bank, in constant 2015 US dollar value from 1978 to 2021. Gini indexes are from
Jiandong Chen, Pu, and Hou 2019 for 1978 to 1980, and for 2002, also from Ravallion and Shaohua Chen 2007 for 1981 to 2001, and
from NBS for 2003 to 2021.
Nonetheless, since the Gini Index in China reached its peak of around 0.5 in 2008, the
trend of income inequality has been restrained over the last 15 years. Some contend that
the rising trend of income inequality has reversed in recent years (Wan, T. Wu, and Y.
Zhang 2018 and Kanbur, Y. Wang, and X. Zhang 2021, for example). One of the most
significant pieces of evidence supporting this argument comes from official reports pub-
lished by the National Bureau of Statistics (NBS) in China. Figure 2.1 illustrates that the
Gini Index has experienced a slow yet consistent decline from 0.49 in 2008 to 0.466 in 2021.
Conversely, Piketty, L. Yang, and Zucman 2019 proposes that income inequality in China
does not display any evident signs of decline, as the income share of the top 10 percent,
the middle 40 percent, and the bottom 50 percent remained stable over the decade pre-
ceding 2015. Furthermore, Ravallion and Shaohua Chen 2022 also claims that the Kuznets
41
growth model (Kuznets 1955) of dual economy transformation with population urban-
ization cannot account for the changes in inequality measures in China dating back to
1981.
In this paper, we participate in the discussion by exploring the evolution of educa-
tion returns in China, using the categorical random coefficient model proposed by Z. Gao
and Pesaran 2023 and a household survey dataset from China Household Income Project
(CHIP) covering 1995, 2002, 2007, 2013, and 2018. We choose to focus on wage income
and changes in education return for two main reasons. First, there is a consensus that the
disparity between urban and rural households has been among the largest contributors to
the overall income inequality in China since 1978 (Sicular, Ximing, and Shi 2007 and X. Liu
2010 for example). Second, wage income is more and more important for rural household
total income composition, and it is also still the largest part of urban household income,
even decreasing gradually from 75% to 60% between 1978 and 2021, a comprehensive un-
derstanding of wage income dispersion can provide insights into the sources of overall
income inequality evolution in the two time periods of 1978 to 2008 and after 2008.
Moreover, the experience of income inequality evolution in China since 1978 has more
in common with the United States and perhaps other Western countries in their earlier
development stages, although China’s rapid economic growth resembles the high-growth
period of its Asian neighbors (J. Zhang 2021). Lemieux 2006a; b also shows that wage
inequality rises in the US during a similar period (1973 to 2005) due to unequal increases in
education returns, especially for postsecondary graduates. Hence it is insightful to explore
the change of heterogeneous education returns in China after 1978 too to make a linkage
with the evolution of wage inequality and income inequality.
Following the setup of a; b and also Heckman, Humphries, and Veramendi 2018 study-
ing the heterogeneity in return to education, in this paper, we use the categorical random
coefficient model proposed by Z. Gao and Pesaran 2023 to explore the evolution of edu-
cation returns in China after the economic reform. We use this model because first, it can
42
provide information on the distribution of the random education return coefficients in ad-
dition to the first and second moments, and second, it can avoid the intrinsic endogeneity
problem in the estimation process of the Lemieux 2006b.
In this paper, we show several findings on the evolution of education return in China.
First, the overall education return was high in 1995 at 6.36% and increased again in 2002,
stayed almost the same in 2007, and dropped in 2013 and 2018, while the dispersion of
education return also increased after 2007. Second, the rural education return was higher
than the urban education return in 1995, but from 2002, the rural mean education return
was lower than the urban mean education returns. The gap shrunk slightly in 2018 but
is still very high. Third, consistent with previous findings, our results show a large post-
secondary premier in education return in the past decade, in both 2013 and 2018, the ed-
ucation return for post-secondary graduates are higher than 14%, while the return to ed-
ucation for individuals who attained high school or below is just 2.15% in 2013 and 1.9%
in 2018. While the education return for high school graduates or lower was higher than
that of postsecondary graduates in 1995 and in 2007 and was also 6.79% in 2002, even
lower than postsecondary graduates. The dispersion of education returns within the high
school or below graduates was also increasing from 2002 to 2007, and to 2013. Fourth,
female education return is always higher than male education returns throughout the pe-
riods, even though both of them first increased from 1995 to 2002, slightly declined in 2007,
and declined largely in 2013 and 2018. The dispersion of both female and male education
returns increased during the periods. In general, the overall education returns show a de-
creasing trend after 2007, as well as estimated using subsamples, including rural/urban
divide, by genders, and by highest educational attainment. The overall dispersion and
also dispersions between rural/urban divide and between postsecondary and high school
or below graduates, and also within-group dispersions all increased, especially for the
recent decade.
43
This paper complements the literature in different strands. First, in the topics about
the rising inequality in China. The rise in income inequality in China since the economic
reform in 1978 has drawn extensive attention from researchers (Gustafsson and Shi 2002,
Benjamin, Brandt, and Giles 2005, Khan and Riskin 2005, Ravallion and Shaohua Chen
2007, Sicular, Ximing, Gustafsson, and Shi 2007, Piketty and Qian 2009, Piketty, L. Yang,
and Zucman 2019, Jiandong Chen, Pu, and Hou 2019, Y. Xie and X. Zhou 2014, Knight
2014, and also reviewed by J. Zhang 2021). As for the recent decade, studies are more
concentrated on the turning around of the Gini Index in China (Kanbur, Y. Wang, and
X. Zhang 2021, Wan, T. Wu, and Y. Zhang 2018, Ravallion and Shaohua Chen 2022, F. Cai
2021). We show that the overall dispersion of education returns in China is increasing
in the time period that we study, as well as the education return dispersion between the
urban-rural divide and the education return dispersion within rural households, and this
can be linked to the recent mild decline of household income inequality, and our finding
is consistent with Ravallion and Shaohua Chen 2022 and Piketty, L. Yang, and Zucman
2019 that there was no convincing sign of turning around and we should not presume the
inequality level in China will mechanically decline.
Secondly, we contribute to the existing literature on education return in China. Byron
and Manaloto 1990, Y. Zhao 1997, T. Li and J. Zhang 1998, J. Zhang, Y. Zhao, Park, and
X. Song 2005, Q. Li, Brauw, Rozelle, and L. Zhang 2005, De Mel, McKenzie, and Woodruff
2008, and Hongbin Li, P. W. Liu, and J. Zhang 2012 assessed education return in China
during its early stages, finding relatively low returns, particularly in rural areas. Later re-
search by Haizheng Li 2003, X. Wu and Y. Xie 2003, J. Zhang, Y. Zhao, Park, and X. Song
2005, Z. Liu 2007, Fang, Eggleston, Rizzo, Rozelle, and Zeckhauser 2012, Knight, Deng,
and S. Li 2017, and M Niaz Asadullah and S. Xiao 2019 focused on the growing impor-
tance of education and human capital as a result of economic reforms. Several studies,
such as those by Hu and Hibel 2014, Mark R. Rosenzweig and J. Zhang 2013, Messinis
2013, X. Wang, Fleisher, Haizheng Li, and S. Li 2014, W. Gao and Smyth 2014, Sch¨ oEmann
44
and Becker 2015, Pi and P. Zhang 2018, and Lin Liu, Paudel, G. Li, and M. Lei 2019, have
attempted to establish a link between heterogeneous education returns and inequalities.
Churchill and Mishra 2018 and X. Ma and Iwasaki 2021 employed meta-analysis tech-
niques to demonstrate the heterogeneity in education return across sectors, urban/rural
areas, gender, hukou, and higher education levels, while Shuxing Chen, L. Zheng, and Y.
Gao 2022 utilized an urban-rural dichotomy model to show how divergences in education
return could lead to differences in educational achievement. More recent studies, such
as those by M. Niaz Asadullah and S. Xiao 2020, Jie Chen and Pastore 2021, and X. Gao
and M. Li 2022, have documented a decline in education returns in China over the past
decade. In this paper, we employ a novel categorical random coefficient model to estimate
both the average and disparity of education returns in the past two decades. Our work
aims to shed light on the relationship between education returns and changes in wage
and income inequality.
The rest of this paper will be organized as the following: the second section will be
background, including the post-reform evolution of income inequality in China, and sources
of inequality; in the third section, we discuss the data source; in the fourth second, we in-
troduce the heterogeneous education return model and the categorical random coefficient
model we use; the fifth section contains the estimate results; and the last section is a con-
clusion.
2.2 Background
Based on evidence from official and nonofficial resources, we notice that income inequality
has been increasing largely in China since 1978. Furthermore, income inequality decompo-
sition also shows between-group gaps and within-group gaps, and also different sources
of inequality and how they were changing over the years, including the rural-urban di-
vide, regional divide, and factor income. Existing studies and administrative data show
45
that the gap between rural and urban household income is the largest in mounting na-
tional income inequality. It was also accompanied by the proportional increase in wage
income within rural households after the economic reform and the huge rural-to-urban
migrant flows. In this paper, we make contributions to the discussion of income inequality
sources and the potential turning point by providing more comprehensive descriptions of
education returns in China overall and also for rural and urban residents separately and
for educational attainment higher than advanced education, including both the dynamics
of mean returns and changes on the dispersion of education returns.
2.2.1 Evolution of Income Inequality in China after 1978
We will show the evolution of income inequality in China after the economic reform in
1978 through both the official information of NBS and also show the inequality levels
measured by previous studies using University-based unofficial social survey projects in-
cluding CHIP, China Family Panel Survey (CFPS), Chinese General Social Survey (CGSS),
China Household Finance Survey (CHFS), and China Labor Force Dynamic Survey (CLDS).
Figure 2.1 shows the evolution of the Gini index in China, as well as the GDP per capita
increase since 1978, especially after 2001 when China joined the WTO. the Gini Index data
is from NBS after 2003, and also from Ravallion and Shaohua Chen 2007 for 1981 to 2001,
as well as from Jiandong Chen, Pu, and Hou 2019 for 1978 to 1980 and for 2002. The
figure shows that first, China’s economic reform and export-oriented economy transfor-
mation achieved huge economic growth. Using constant 2015 US dollar values, the GDP
per capita in China grew from $381.1 in 1978 to $2193.9 in 2000 continuously and stably,
mostly due to economic reform in both rural and urban areas. Then after 2001, the growth
rates significantly increased and rapidly increased from $2359.6 in 2001 to$11188.3 in 2021.
At the same time, the general household income inequality also increased drastically even
before 2001 and stayed high until the end of 2010s. Figure 2.1 shows that after 1978, the
Gini Index first declined from a high point of 0.317 in 1980 to 0.289 in 1985, this was due to
46
the initial economics reform being concentrated in rural areas, and the agricultural market
reform increasing rural households’ income. Then, after the reform’s focus moved to the
urban economy in 1985, the Gini Index increased from a relatively low level of 0.289 in
1985 to a high level of 0.491 in 2008, with some bouncing around periods between 1994
and 1995.
However, after around 2008, the rise of the Gini Index has been stalled and even showed
some signs of a declining trend. The Gini Index has dropped from its highest point of
close to 0.5 in 2008 to a low point of 0.466 in 2021. Some recent studies claimed that this is
a sign that China has already passed the turning point (Wan, T. Wu, and Y. Zhang 2018,
Kanbur, Y. Wang, and X. Zhang 2021), while some other researchers show concerns about
the claimed turning point. Piketty, L. Yang, and Zucman 2019 mentioned income under-
reporting from the rich and wealth inequality; Ravallion and Shaohua Chen 2022 shows
that the structural transformation described in the Kuznets model has little explanatory
power for the inequality measures change in China for the recent trends. We will provide
more evidence on the dispersion of education returns and also show that the evolution of
education returns also could not support the turning point optimism either.
47
Figure 2.2: Gini Index in China Since 1978 from Various Sources from 2002 to 2015
Note: The figure is from J. Zhang 2021. CHIP stands for China Household Income Project, CFPS stands for China Family Panel Survey,
CGSS stands for Chinese General Social Survey, CHFS stands for China Household Finance Survey, and CLDS stands for China Labor
Force Dynamic Survey.
Alternatively, besides the official information from the NBS, there are some university-
based social survey projects since 2010, including CFPS, CGSS, CHFS, and CLDS. Exten-
sive studies have been done using those household survey datasets to measure the national
income Gini Index. Figure 2.2 is a graph in and it summarized recent attempts at measur-
ing inequality using these survey datasets. Consistent with the official dataset of NBS and
the CHIP projects which is a project cooperated with NBS, the Gini index measured by
the unofficial data sources also shows that inequality was always at a very high level, and
kept increasing until the end of the 2000s, while it also shows some sign of being steady
and even slightly decreasing in the recent decade.
2.2.2 Rural-Urban Gaps and Income Sources
The rural-urban gaps have been prolongedly considered the largest contributor to income
inequality in China since 1978. Atinc 1997 claims that the urban-rural income disparity
contributed three-fourths of the overall income inequality in the 1990s. Similarly, D. T.
48
Yang 1999 uses information from two provinces showing that the urban-rural income dif-
ferentials contributed 75% of total income inequality in Sichuan, and 40% in Jiangsu us-
ing the Theil coefficient decomposition. Sicular, Ximing, and Shi 2007 use more detailed
survey datasets on the China Household Income Project (CHIP) and show that the rural-
urban 41% in 1995 and 46% in 2002; These findings are consistent with X. Liu 2010 who
decomposes China’s income inequality and find that the contribution of the urban-rural
gap to overall income inequality in recent China had markedly increased from 57.98% in
1997 to 72.84% in 2006 using the China Statistical Yearbook from the National Bureau of
Statistics (NBS). This happens mostly because of the hukou system and differential public
service provided based on hukou as well as urban-biased policies (D. T. Yang 1999). Fig-
ure 2.3 shows the evolution of the Gini Index and the urban-to-rural household income
ratio in China since 1978. It shows that the urban-to-rural household income ratio and the
Gini Index in China were highly correlated and moved almost homogeneously. Similar
to the Gini Index, the income ratio decreased from 2.56 in 1978 to 1.84 in 1984, due to the
initial rural economic reform, and after the reform’s focus moved to the urban economy
and further openness, the ratio quickly increased to 3.14 in 2007, and it was also the year
when the Gini Index reached the highest point. Also similarly, after 2008, the urban-to-
rural household income ratio drop slowly and persistently to 2.5 in 2021, which was the
same level as the level in 1998.
49
Figure 2.3: Urban-to-rural Income Ratio and Gini Index in China Since 1978
Note: Average urban household income and average rural household income are from yearly UHS and RHS projects of NBS, 1978 to
2021. Gini indexes are from Jiandong Chen, Pu, and Hou 2019 for 1978 to 1980, and for 2002, also from Ravallion and Shaohua Chen
2007 for 1981 to 2001, and from NBS for 2003 to 2021.
It is worth noticing that before 2007, the correlation between the Gini Index and the
urban-to-rural income ratio was much higher than in the recent period when both of them
kept decreasing. In 2021, the ratio was back to the level in 1998, however, the Gini Index
in 2021 (0.466) is still significantly higher than the index in 1998 (0.403). We claim that
this happens because of the increasing dispersion of education returns.
50
(a) Rural
(b) Urban
Figure 2.4: Rural and Urban Household Income Sources in China from 1998 to 2022
Note: Household average income and four different sources of income information are all from yearly UHS and RHS projects of NBS,
1998 to 2022. Four income sources include income from wages and salaries, net business income, net income from property, and net
income from the transfer.
51
Subfigures 2.4a and 2.4b show the evolution of income sources for both rural and ur-
ban households from 1998 until 2022. Based on the income source categories of the NBS
survey, household income is divided into four sources, including income from wages and
salaries, net business income, net income from property, and net income from transfers.
The net business income includes the net business operating income for urban households
and agricultural income for rural households, and the transfer income contains both public
transfers (pensions, for example) and private transfers (remittances, for example). Figure
2.4a shows that for rural households, there is a persistent trend that the income from the
agricultural source kept declining, from the highest point of 67.8% in 1998 to only 34.6%
in 2022, at the same time, the proportion of wage income increased from 26.3% to 41.9 %
in 2022, and the part of transfer income rose from 4.5% to 20.9% in 2022. Even though
the increase in transfer income contribution was also a result of government subsidies
and welfare-inclusive policies, the rising transfer income is mainly a result of rising re-
mittances due to the huge rural-to-urban migrant flows. Alongside the rising proportion
of wage income, the contribution of salary income is roughly more than half of the total
rural household income now. On the contrary, the proportion of wage income in urban
household income decreased from 75.2% to just 60% in 2022, while income is more from
the business operation and property. However, the proportion of wage income is still very
high for urban households and stayed stable in the past decade.
2.3 Data Source
The primary data source for this study is the China Household Income Project (CHIP)
survey dataset, utilizing information from the years 1995, 2002, 2007, 2013, and 2018, as a
result of data availability constraints. The NBS survey data itself is not open to researchers,
and among all university-based household survey datasets, we chose CHIP because first,
52
it is a cross-sectional survey so it fits our estimation model best; second, it has both de-
tailed income information from different sources, and detailed work time; third, the CHIP
dataset is the only available survey containing enough waves before 2010 so it can be used
to make a meaningful comparison between the period from 1978 until 2010 and the time
period afterward.
CHIP was inaugurated in 1988 as a joint venture between the National Bureau of Statis-
tics (NBS) and the Chinese Academy of Social Science. It was established with the spe-
cific objective of examining income inequality in China following the commencement of
economic reforms in 1978. The inaugural survey was conducted between March and June
1989, encompassing 31,827 family members from 9,007 urban households, and 51,352 fam-
ily members from 10,258 rural households across 28 provinces, offering comprehensive
data on income, consumption, and demographic characteristics. As a collaborative ef-
fort with the NBS, all households were subsampled from the Urban Household Survey
(UHS) with 34,945 households and Rural Household Survey (RHS) with 67,186 house-
holds, both collected by China’s National Bureau of Statistics (NBS). Following the first
round in 1989, seven additional waves of income surveys were conducted in 1995, 1999,
2002, 2007, 2008, 2013, and 2018. Excluding 1999, all surveys encompassed households
from both rural and urban communities, collecting economic and other information using
consistent standards. Furthermore, from 2002 onwards, the dataset incorporated rural-to-
urban migrants as a distinct category of households, which serves as an additional ad-
vantage of this dataset. The survey waves of 2007 and 2008 are part of the Longitudinal
Survey on Rural-Urban Migration in China (RUMiC), a collaborative project between the
Australian National University, Beijing Normal University, and the Institute for the Study
of Labor (IZA). The RUMiC project contains two waves and tracked the same group of
households in both 2008 and 2009
3
.
3
The publicly available dataset of RUMiC is called RUMiC 2008 and RUMiC 2009, corresponding to the
survey year, while the respective CHIP dataset is called CHIP2007 and CHIP2008, corresponding to the year
of which the income and consumption information was surveyed. Akg¨ uc ¸, Giulietti, and Zimmermann 2013
53
In this study, we employ only the 1995, 2002, 2007, 2013, and 2018 CHIP waves, as de-
tailed work time information is unavailable for the 1988 and 1999 waves, precluding the
generation of individual wage rates for primary occupations. Due to the cross-sectional
perspective of the specification we use in this paper, we also exclude CHIP2008. To cal-
culate the wage rate, individual total annual income encompasses income from primary
occupations and all other part-time positions, including salary income and net operating
income for urban households, as well as salary income and non-agricultural operating
income for rural households. Total work hours, either annual or monthly, account for
the total work hours of the primary job and all other part-time positions, incorporating
non-agricultural self-employment. To ensure the comparability of wage rate information
across various years, we also adjust wage rates before 2018 using the annual GDP defla-
tor rate of each respective year concerning 2018
4
. Experience is calculated as age minus
years of education minus six, and additional control variables employed in the analysis
include gender, ethnicity, marital status, and employer type (public sector employment
encompasses official government institutes, non-profit public institutes, and state-owned
enterprises (SOEs)).
We present a summary statistics table illustrating the dynamics of primary indicators
among CHIP samples. In Table 2.1, we display the mean values and standard deviations
of primary variables for each year, encompassing wage rate, age, urban indicator, gen-
der, marital status, ethnicity, education year, experience, and public sector employment
indicator. Notably, even after adjusting for inflation, wage rates experienced a significant
increase from 1995 to 2018, particularly following China’s accession to the World Trade
provides a review of the RUMiC project’s structure and techniques, as well as existing studies based on
RUMiC data. Also, see Kong 2010
4
The inflation indicators are from the World Bank, as measured by the annual growth rate of the GDP
implicit deflator shows the rate of price change in the economy as a whole. The GDP implicit deflator is the
ratio of GDP in the current local currency to GDP in constant local currency. https://data.worldbank.o
rg/indicator/NY.GDP.DEFL.KD.ZG?view=chart
54
1995 2002 2007 2013 2018 Total
Wage 5.44 6.89 13.06 17.67 25.38 15.28
(4.17) (8.89) (19.47) (44.27) (52.30) (35.87)
Age 37.45 37.95 34.89 38.37 40.88 38.16
(9.97) (10.76) (11.28) (11.64) (11.74) (11.44)
Urban 0.91 0.59 0.58 0.41 0.56 0.59
(0.29) (0.49) (0.49) (0.49) (0.50) (0.49)
Gender 0.55 0.63 0.61 0.62 0.61 0.61
(0.50) (0.48) (0.49) (0.48) (0.49) (0.49)
Marital Status 0.84 0.83 0.73 0.80 0.84 0.81
(0.37) (0.38) (0.44) (0.40) (0.36) (0.39)
Han=1 0.96 0.94 0.99 0.95 0.94 0.95
(0.20) (0.25) (0.10) (0.22) (0.23) (0.21)
Education years 10.47 9.48 9.78 10.07 10.21 9.97
(3.06) (3.27) (3.14) (3.43) (3.58) (3.36)
Experience(estimated) 20.97 22.45 19.00 22.26 24.58 22.13
(10.32) (11.45) (12.05) (13.05) (13.54) (12.54)
Work in public sector 0.90 0.33 0.30 0.24 0.19 0.33
(0.29) (0.47) (0.46) (0.43) (0.39) (0.47)
Observations 12593 23422 23348 17947 31553 108863
Table 2.1: Summary Statistics
Note: All data are from the China Household Income Project (CHIP), 1995, 2002, 2007, 2013, and 2018. Wage rates are calculated
using annual salary and business operating income divided by total work hours for urban residents and salary income plus
non-agricultural income divided by total work hours for rural residents. Marital status equals one if the individual is within a
marriage or cohabitation status. Experience is calculated by age fewer education years and 6. Observations are the total amount of
individuals that worked during the survey year and have a valid positive wage rate at each wave or in total.
55
Organization (WTO) in 2002. Conversely, self-reported education year attainments re-
mained relatively stable between 1995 and 2018, ranging from 9.5 to 10.5 years, suggest-
ing a concurrent increase in education returns. It is also noteworthy that in 1995, more
than 90% of employed wage earners originated from urban communities and were em-
ployed in the public sector. In 2002, this proportion dropped substantially to one-third
and continued to decline annually until 2018.
2.4 Empirical Strategy
In his seminal contribution, Mincer 1974 models the logarithm of earnings as a function
of years of education and years of potential labor market experience, which can be written
in a generic form,
logwage
i
=α
i
+β
i
edu
i
+ϕ (z
i
) +ε
i
, (2.1)
as in Heckman, Humphries, and Veramendi 2018, Equation 1, wherez
i
includes the labor
market experience and other relevant control variables. The above wage equation, also
known as the ”Mincer equation”, has become the workhorse of the empirical works on
estimating the return to education. In the most widely used specification of the Mincer
equation (2.1),
ϕ (z
i
) =ρ
1
exper
i
+ρ
2
exper
2
i
+ ˜ z
′
i
˜ γ,
where ˜ z
i
is the vector of control variables other than potential labor market experience.
As the literature has witnessed in the past decades, there has been a growing interest in
heterogeneity in return to education. Accordingly, it is important to allow the parameters
in the Mincer equation to vary across individuals. In recent work, Z. Gao and Pesaran 2023
consider a linear regression model with categorical random coefficients, which allows α
i
56
andβ
i
in (2.1) to differ across individuals. The model restricts the return to education to
follow a categorical distribution,
β
i
=
b
1
, w.p. π
1
b
2
, w.p. π
2
.
.
.
.
.
.
b
K
, w.p. π
K
,
where w.p. denote “with probability”,π
k
∈ (0,1), and
P
K
k=1
π
k
= 1. Under regularity con-
ditions, both the moments ofβ
i
, mean and variance for examples, and the distributional
parameters, b
k
and π
k
can be identified and estimated by a generalized method of mo-
ments estimator relying on moment conditions. The estimator is shown to be consistent
and asymptotic normal.
In , the categorical random coefficient model is applied to study the heterogeneous re-
turn to education in the U.S. from 1973 to 2003, following a similar setup as in Lemieux
2006a and b, which first studied the issue with a variance decomposition approach. Com-
pared to the b, the categorical random coefficient model can provide information on the
distribution of the random coefficient β
i
in addition to the first and second moments, and
it avoids the intrinsic endogeneity problem in the estimation procedure as in b
5
.
Following Z. Gao and Pesaran 2023, we set up the categorical coefficient model for a
repeat cross-sectional sample where the time variation in the parameters is index byt,
logwage
it
=α
it
+β
it
edu
it
+ρ
1t
exper
it
+ρ
2t
exper
2
it
+ ˜ z
′
it
˜ γ
t
+ε
it
, (2.2)
5
The details are relegated to Appendix F.
57
where the return to education follows the categorical distribution,
6
β
it
=
b
tL
w.p. π
t
,
b
tH
w.p. 1−π
t
,
and ˜ z
it
includes relevant control variables such as gender and marital status. α
it
=α
t
+δ
it
where δ
it
is mean 0 random variable assumed to be distributed independently of edu
it
andz
it
=
exper
it
,exper
2
it
, ˜ z
′
t
′
. Letu
it
=ε
it
+δ
it
, and write (2.2) as
logwage
it
=α
t
+β
it
edu
it
+ρ
1t
exper
it
+ρ
2t
exper
2
it
+ ˜ z
′
it
˜ γ
t
+u
it
. (2.3)
We estimate 2.3 for each sub-sample and sample period.
2.5 Results
Using CHIP data of multiple waves in 1995, 1999, 2002, 2007, 2013, and 2018, and the
categorical coefficient model in , we estimate the heterogeneous education return model
2.
2.5.1 Evolution of Education Returns in China
First, we examine the overall evolution of education return in China, along with the changes
in distribution. Table 2.2 presents our estimated results of overall education returns on the
wage rate. The third column represents the estimatedπ
t
in the setup of Equation 2, signify-
ing the estimated proportion of individuals in the low return category. β
L
andβ
H
denote
estimates of the low and high categorical returns, whileE(β
i
) ands.d.(β
i
) are the mean
and standard deviation of returns calculated using the previous estimates. E(βi
GMM
)
6
In practice, we set the number of categoriesK =2 for simplicity. As explained in , the method delivers
credible results for smallK. Note that the estimates for moments ofβ
i
does not dependent on the choice of
K.
58
ands.d.(βi
GMM
) represent the mean and standard deviation ofβ
i
estimated directly from
the moment conditions and the GMM estimation method. We primarily concentrate on
the estimated mean value and standard deviation of education return using GMM for the
remainder of this section.
Year p β
L
β
H
E(β
i
) s.d.(β
i
) E(β
GMM
i
) s.d.(β
GMM
i
) N
1995 All 0.4714 0.0612 0.0612 0.0612 0.0000 0.0636 n/a 11582
2002 All 0.4271 0.0700 0.0700 0.0700 0.0000 0.0704 n/a 21995
2007 All 0.0072 -0.1786 0.0735 0.0717 0.0213 0.0689 0.0194 13199
2013 All 0.4512 0.0084 0.0511 0.0318 0.0212 0.0314 0.0209 17800
2018 All 0.3981 0.0328 0.0330 0.0329 0.0001 0.0301 n/a 30886
Table 2.2: Estimates of the Distribution of the Education Return
Note: This table reports the estimates of the distribution ofβi with the quadratic in experience specification 2, using S =4 order
moments ofedu
i
. ”All” stands for all individuals working by the end of the specific survey year and reports both valid total salaries
and business operating or non-agricultural business operating income and total work hours for the specific years. Controls in the
estimation contain gender, urban/rural, marital status, minzu (ethnicity minority), and public (whether employed by the
government, official social organizations, or state-owned enterprises (SOEs)).
′′
n/a
′′
is inserted when the estimates show
homogeneity ofβ
i
and standard deviations could not be estimated. All data are from the China Household Income Project (CHIP) in
the years 1995, 2002, 2007, 2013, and 2018.
In line with previous measures of education return in China since the economic reform,
which indicate a significant increase in education return, our initial estimated mean return
to education in 1995 is 6.36%. In 2002, it further increased to 7.04% and slightly declined to
6.89% in 2007, remaining high. However, in 2013, the mean education return dropped by
more than half to 3.14% and further declined to 3.01% in 2018. This decline in education
returns in recent years is also observed by Jie Chen and Pastore 2021, M. Niaz Asadullah
and S. Xiao 2020, and X. Gao and M. Li 2022, with our estimates exhibiting a larger extent
of decline
7
.
The GMM estimates of β
i
exhibit homogeneity in 1995 and 2002, while indicating
higher heterogeneity in 2007 and 2013. The dispersion of education return in 2013 sur-
passes that of 2007, suggesting increased dispersion of return to education in the past
decade. In 2018, the national estimate also displays no evident heterogeneity.
7
M Niaz Asadullah and S. Xiao 2019, for example, found that education return in urban locations de-
creased from 10.4% to 8.3%, from 9.9% to 7.8% for coastal regions, and from 7.8% to 6.3% for women.
59
2.5.2 Distribution of the Education Return by Rural/Urban Divide
We present estimated results utilizing subsamples segregated by rural and urban resi-
dents.Table 2.3 reveals that in 1995, the rural education return of 7.78% surpassed the ur-
ban return of 6.09%. However, this pattern does not persist in subsequent periods (2002,
2013, and 2018). In 2002, rural education returns plummeted to 4.19%, while urban returns
soared to 8.86%. By 2007, the survey exclusively captured urban households, yielding a
mean return of 6.89% in cities. In 2013, the mean rural education return further dimin-
ished to 0.84% and urban returns reached 8.21%. Lastly, in 2018, rural education returns
remained low at 1.35% while urban returns decreased to 4.7%.
This divergence between urban and rural households indicates that the decline in over-
all education returns during 2013 and 2018 is predominantly attributable to the plummet-
ing rural education returns. Despite the urban education return also dropping signifi-
cantly in 2018, it increased in 2013 compared to 2007. The widening gap in education re-
turns could exacerbate income inequality in the rural-urban divide, particularly as wage
income becomes increasingly vital for rural households and comprises over 60% of urban
household income. Furthermore, the dwindling rural education returns may influence
household education investment decisions, potentially exacerbating intergenerational mo-
bility concerns.
With regard to within-group dispersion, albeit incomplete evidence, the education re-
turn dispersion decreased for both rural and urban areas from 2002 to 2013. This under-
scores that the between-group difference may become increasingly crucial for the rural-
urban divide when compared to within-rural or within-urban gaps.
2.5.3 Distribution of the Education Return by Highest Education Attained
Consistent with the postsecondary premium on education return identified by Lemieux
2006b in the US, we observe a substantial postsecondary premium for postsecondary grad-
uates in China over the past decade. Table 2.4 illustrates that in 1995, the education return
60
Year p β
L
β
H
E(β
i
) s.d.(β
i
) E(β
GMM
i
) s.d.(β
GMM
i
) N
1995 All 0.4714 0.0612 0.0612 0.0612 0.0000 0.0636 n/a 11582
Rural 0.4227 0.0625 0.0625 0.0625 0.0000 0.0778 n/a 979
Urban 0.4366 0.0585 0.0585 0.0585 0.0000 0.0609 n/a 10603
2002 All 0.4271 0.0700 0.0700 0.0700 0.0000 0.0704 n/a 21995
Rural 0.1092 -0.0595 0.0543 0.0419 0.0355 0.0419 0.0355 8927
Urban 0.4627 0.0894 0.0895 0.0894 0.0000 0.0886 n/a 13068
2007 All 0.0072 -0.1786 0.0735 0.0717 0.0213 0.0689 0.0194 13199
Rural n/a n/a n/a n/a n/a n/a n/a
Urban 0.0072 -0.1786 0.0735 0.0717 0.0213 0.0689 0.0194 13199
2013 All 0.4512 0.0084 0.0511 0.0318 0.0212 0.0314 0.0209 17800
Rural 0.9656 0.0063 0.0782 0.0087 0.0131 0.0084 0.0120 10415
Urban 0.0000 -1.4364 0.0811 0.0811 0.0029 0.0821 0.0028 7385
2018 All 0.3981 0.0328 0.0330 0.0329 0.0001 0.0301 n/a 30886
Rural 0.4577 0.0169 0.0169 0.0169 0.0000 0.0135 n/a 13515
Urban 0.0002 -0.6951 0.0477 0.0476 0.0101 0.0470 0.0158 17371
Table 2.3: Estimates of the Distribution of the Education Return by Rural/Urban Divide
Note: This table reports the estimates of the distribution ofβi with the quadratic in experience specification 2, using S =4 order
moments ofedu
i
. ”All” stands for all individuals working by the end of the specific survey year and also reports both valid total
salaries and business operating or non-agricultural business operating income and total work hours for the specific years. Results
using subsamples in urban and rural communities are all reported. Controls in the estimation contain gender, urban/rural, marital
status, minzu (ethnicity minority), and public (whether employed by the government, official social organizations, or state-owned
enterprises (SOEs)). For estimates using subsamples, controls do not include the indicator of the urban/rural divide.
′′
n/a
′′
is
inserted when the estimates show homogeneity ofβ
i
and standard deviations could not be estimated. Estimate results for rural
subsamples in 2007 are all unavailable because CHIP 2007 only surveyed urban households at that wave year. All data are from the
China Household Income Project (CHIP) in the years 1995, 2002, 2007, 2013, and 2018.
61
for high school or below employees was 7.34%, while the return for postsecondary grad-
uates was a mere 4.53%. In 2002, the education return for high school or below graduates
rose to 6.79%, with postsecondary graduates experiencing a return of 9.47%. By 2007, the
education return further increased to 7.97% for high school or below graduates and 2.29%
for college graduates. In 2013 and 2018, postsecondary education returns surged to 14.62%
and 14.52%, respectively, while the education return for high school or below graduates
dwindled to 2.15% and 1.12%.
Year p β
L
β
H
E(β
i
) s.d.(β
i
) E(β
GMM
i
) s.d.(β
GMM
i
) N
1995 All 0.4714 0.0612 0.0612 0.0612 0.0000 0.0636 n/a 11582
High School or Below 0.4042 0.0691 0.0692 0.0692 0.0000 0.0734 n/a 8901
Postsecondary 0.4159 0.0453 0.0453 0.0453 0.0000 0.0453 n/a 2681
2002 All 0.4271 0.0700 0.0700 0.0700 0.0000 0.0704 n/a 21995
High School or Below 0.0000 -0.0470 0.0677 0.0677 0.0000 0.0679 0.0161 18595
Postsecondary 0.4734 0.0982 0.0982 0.0982 0.0000 0.0947 n/a 3400
2007 All 0.0072 -0.1786 0.0735 0.0717 0.0213 0.0689 0.0194 13199
High School or Below 0.0057 -0.1629 0.0807 0.0793 0.0183 0.0797 0.0185 10345
Postsecondary 0.4711 0.0201 0.0201 0.0201 0.0000 0.0229 n/a 2854
2013 All 0.4512 0.0084 0.0511 0.0318 0.0212 0.0314 0.0209 17800
High School or Below 0.9436 0.0174 0.1239 0.0234 0.0246 0.0215 0.0212 14104
Postsecondary 0.2168 0.1011 0.1600 0.1473 0.0243 0.1462 0.0219 3696
2018 All 0.3981 0.0328 0.0330 0.0329 0.0001 0.0301 n/a 30886
High School or Below 0.3684 0.0208 0.0267 0.0245 0.0028 0.0190 n/a 23581
Postsecondary 0.4565 0.1331 0.1555 0.1452 0.0112 0.1452 n/a 7305
Table 2.4: Estimates of the Distribution of the Education Return by Highest Education
Attained
Note: This table reports the estimates of the distribution ofβi with the quadratic in experience specification 2, using S =4 order
moments ofedu
i
. ”All” stands for all individuals working by the end of the specific survey year and also reports both valid total
salaries and business operating or non-agricultural business operating income and total work hours for the specific years. Results
using subsamples divided by the highest education attained higher than post-secondary and high school or less are also reported.
Controls in the estimation contain gender, urban/rural, marital status, minzu (ethnicity minority), and a dummy for postsecondary
education and public (whether employed by the government, official social organizations, or state-owned enterprises (SOEs)). For
estimates using subsamples, controls do not include the indicator of postsecondary education.
′′
n/a
′′
is inserted when the estimates
show homogeneity ofβ
i
and standard deviations could not be estimated. All data are from the China Household Income Project
(CHIP) in the years 1995, 2002, 2007, 2013, and 2018.
Given that the sample size of individuals with postsecondary education or higher was
insufficient for all years except 2018, these estimates may be less informative. Nonethe-
less, the results clearly illustrate a growing divergence in education returns between post-
secondary graduates and high school or below graduates. The dispersion of education
returns for high school or below graduates continuously increased from 2002 to 2007 and
62
further to 2013, reflecting the overall dispersion trend. This indicates that the education
return distribution for high school and below has become flatter and shifted to the left.
2.5.4 Distribution of the Education Return by Gender
We further present estimates of education returns using subsamples segregated by gender.
Over the sample period from 1995 to 2018, female education returns consistently surpass
male education returns, with the gap being more pronounced in the recent decade. Table
2.5 indicates that in 1995, the average female education return was 6.96%, while the male
average education return was 6.06%, lower than the female return but with a relatively
small difference. In 2002, the female education return rose to 8.35%, while the male return
remained at the same level as in 1995, at 6.09%, widening the gap. In 2007, the female
return decreased to 6.8%, and the male education return grew to 6.79%, becoming nearly
equivalent to female returns. In 2013, both female and male returns declined, reaching
4.16% and 2.23% respectively, with a considerably larger gap. In 2018, the female average
education return further dropped to 3.52%, while the male education return increased to
2.43%, remaining significantly lower than the female education return. Concerning the
dispersion of education returns, female dispersion expanded from 2002 to 2007 and again
from 2007 to 2013.
2.6 Conclusion
Over the past three decades, China has experienced an economic miracle, accompanied
by a surge in income inequality to a level among the highest globally. Around 2010, there
appeared to be signs of stabilization and even a declining trend in China’s Gini Index.
In this paper, we attempt to enhance our understanding of China’s future income in-
equality. Using survey data from the China Household Income Project (CHIP) for 1995,
2002, 2007, 2013, and 2018, we employ a categorical random coefficient model to estimate a
63
Year p β
L
β
H
E(β
i
) s.d.(β
i
) E(β
GMM
i
) s.d.(β
GMM
i
) N
1995 All 0.4714 0.0612 0.0612 0.0612 0.0000 0.0636 n/a 11582
Female 0.4634 0.0673 0.0673 0.0673 0.0000 0.0696 n/a 5248
Male 0.4626 0.0571 0.0571 0.0571 0.0000 0.0606 n/a 6334
2002 All 0.4271 0.0700 0.0700 0.0700 0.0000 0.0704 n/a 21995
Female 0.0000 -0.8499 0.0826 0.0826 0.0053 0.0835 0.0059 8098
Male 0.4157 0.0603 0.0603 0.0603 0.0000 0.0609 n/a 13897
2007 All 0.0072 -0.1786 0.0735 0.0717 0.0213 0.0689 0.0194 13199
Female 0.0099 -0.1635 0.0726 0.0703 0.0234 0.0680 0.0215 5418
Male 0.0045 -0.2110 0.0724 0.0711 0.0190 0.0679 0.0180 7781
2013 All 0.4512 0.0084 0.0511 0.0318 0.0212 0.0314 0.0209 17800
Female 0.7521 0.0262 0.0912 0.0423 0.0281 0.0416 0.0266 6709
Male 0.0019 -0.2734 0.0227 0.0222 0.0131 0.0223 0.0130 11091
2018 All 0.3981 0.0328 0.0330 0.0329 0.0001 0.0301 n/a 30886
Female 0.4572 0.0381 0.0382 0.0382 0.0001 0.0352 n/a 12008
Male 0.3573 0.0274 0.0274 0.0274 0.0000 0.0243 n/a 18878
Table 2.5: Estimates of the Distribution of the Education Return by Gender
Note: This table reports the estimates of the distribution ofβi with the quadratic in experience specification 2, using S =4 order
moments ofedu
i
. ”All” stands for all individuals working by the end of the specific survey year and also reports both valid total
salaries and business operating or non-agricultural business operating income and total work hours for the specific years. Results
using subsamples divided by gender are also reported. Controls in the estimation contain gender, urban/rural, marital status, minzu
(ethnicity minority), and a dummy for postsecondary education and public (whether employed by the government, official social
organizations, or state-owned enterprises (SOEs)). For estimates using subsamples, controls do not include gender.
′′
n/a
′′
is
inserted when the estimates show homogeneity ofβ
i
and standard deviations could not be estimated. All data are from the China
Household Income Project (CHIP) in the years 1995, 2002, 2007, 2013, and 2018.
64
heterogeneous education return equation, providing information on both the mean value
of education returns overall and by different subsamples, as well as changes in the distri-
bution of education returns in China from 1995 to 2002.
Our findings indicate that the overall education return increased slightly from 1995 to
2002, remained stable until 2007, and then declined substantially in 2013 and 2018, with
the dispersion also increasing from 2007 to 2013. Examining subsamples for rural and
urban households, we observe that although rural education returns were higher than
urban returns in 1995, urban education returns have consistently surpassed rural returns
since then, with the gap widening in the past decade. The decline in education returns was
more pronounced among rural individuals. Similarly, in 1995, there was no postsecondary
premium and high school or lower graduates had higher education returns. However, the
education return of postsecondary graduates was consistently higher than that of high
school or lower graduates, except in 2007, with these gaps considerably larger in 2013
and 2018. Furthermore, the within-group education return dispersion increased for high
school or below graduates from 2002 to 2007 and 2013. Regarding subsamples separated
by gender, females consistently exhibited higher education returns in our estimates, with
the gap being moderate before 2007 and larger in 2013 and 2018. Moreover, the dispersion
of education returns among females increased from 2002 to 2007 and 2013. In summary,
we demonstrate that education returns in China have become more dispersed, particularly
in the past decade, with this expansion of dispersion encompassing general measures, the
urban/rural divide, the highest education divide, the gender divide, and within those
groups.
Our estimates reveal that the decline in education returns and the expansion of edu-
cation return dispersion in multiple ways between and within groups are occurring con-
currently, especially in the past decade. Given that rural/urban inequality remains the
largest contributor to income inequality in China, and wage income is becoming increas-
ingly important among all income sources, the trend of education returns could exert more
65
pressure on income inequality. Without further measures to support rural areas and pro-
mote regional equity, there is no guarantee that the decline in income inequality in China
will occur naturally.
66
Chapter 3
High-Speed Network and Self-Employment —Evidence
from CFPS
3.1 Introduction
In many developing countries, the labor market often lacks sufficient formal sector jobs,
particularly for young people. However, self-employment and entrepreneurship can serve
as profitable alternative options (De Mel, McKenzie, and Woodruff 2008). For low-educated
females from disadvantaged backgrounds, self-employment may be their only career choice.
Furthermore, compared to formal wage positions in sweatshops, self-employment can
lead to better mental and physical health conditions (Blattman and Dercon 2018).
Entrepreneurship and small businesses are therefore considered more effective ways
to address unemployment issues and increase income. Typically, small businesses in de-
veloping countries are quite small in size, with very few employees, often none. There
has been extensive research on the constraints faced by these small businesses and how to
help them grow through intervention. Credit constraints are commonly identified as sig-
nificant barriers (Banerjee and Duflo 2014), along with saving constraints and managerial
skills. To support entrepreneurs in developing countries in establishing or expanding their
small businesses, numerous intervention programs have been conducted recently, focus-
ing on various aspects such as labor drops (De Mel, McKenzie, and Woodruff 2019), cash
67
drops (Blattman, Green, Jamison, Lehmann, and Annan 2016), business training (Mcken-
zie and Puerto 2017), technique drops (Atkin, Chaudhry, Chaudry, Khandelwal, and Ver-
hoogen 2017), and business plan competitions (McKenzie 2017). The evaluation results of
these interventions are mixed, with some finding significant and even sustained impacts,
but many discovering that the effects on female entrepreneurs are not persistent (Blattman
and Dercon 2018). Furthermore, research has shown that women in business face more
obstacles than men, such as gender-based demand constraints (Hardy and Kagy 2020).
Even for self-employed income, females earn less than males, which cannot be explained
by the extensive firm- or owner-level characteristics (2018). This highlights the importance
of considering heterogeneity in small business development.
This paper will examine the relationship between technological development and an
individual’s self-employment choice and income growth. Over the past two decades,
we have witnessed significant technological advancements, particularly the widespread
adoption of high-speed Internet. Given the labor market structure and skill distribution,
new technology could lead to skill-biased technological change, which in turn may result
in both income growth and increased inequality. Hjort and Tian 2021 provides a review
of the economic impacts of Internet connectivity on firms, workers, and consumers in de-
veloping countries. Specifically, Dolfen, Einav, Klenow, Klopack, J. D. Levin, L. Levin,
et al. 2023, Luo and C. Niu 2019, and Couture, Faber, Gu, and Lizhi Liu 2021 investigate
the effects of e-commerce on welfare; Bhuller, Ferraro, Kostøl, and Vigtel 2023, Geraci,
Nardotto, Reggiani, and Sabatini 2022, Balgobin and Dubus 2022, Zuo 2021, and Hjort
and Poulsen 2019 study the impact of the Internet on labor market dynamics and employ-
ment. M. Yang, S. Zheng, and L. Zhou 2022, Lin, C. Ma, Y. Sun, and Y. Xu 2021, and X.
Xu, Watts, and Reed 2019 explore the connection between Internet usage and innovation,
while M. Xie, Ding, Xia, Guo, Jiaofeng Pan, and H. Wang 2021, Shiyi Chen, W. Liu, and
H. Song 2020, and Cusolito, Lederman, and Pe˜ na 2020 examine how Internet connectivity
affects a firm’s productivity and factor demands in general. In terms of the impact of the
68
Internet on an individual’s choices and benefits, L. Chen and W. Liu 2022 links Internet
access to body weights; G. Niu, Jin, Q. Wang, and Yang Zhou 2022, Leng 2022, and Viollaz
and Winkler 2022 also demonstrate that access to the Internet can be advantageous for
disadvantaged groups of individuals.
Within the past decade, China has experienced a large increase in Internet speed, for
both fixed internet and mobile internet. The national weighted average download speed
increased from 2.93Mbit/s in June 2013
1
to 35.46Mbit/s in the second quarter of 2019
2
. Taking advantage of the rapid growth of internet speed, there are several new on-line
business platforms introduced within the past few years in China. Besides the traditional
B2C (Business to Customer) shopping websites like Taobao.com and Tmall.com, there are
more websites or mobile phone applications like Weidian (A shopping APP that is mostly
connected to the largest social media application in China), and Xianyu (online secured
transaction flea market) and many subdivided commodity categories platforms or shop-
ping experience sharing websites like cosmetic or sneakers. Besides those, short video-
sharing applications have also attracted much attention, from both home and abroad, like
Tiktok. Famous Tiktokers can cooperate with businesses and introduce huge consumer
attention. Most of the business owners and shoppers of these websites or mobile applica-
tions are young females and a large proportion of them are doing this as their main job. I
will describe and explain the general impacts of Internet development in the next section.
With the help of technological development and new platforms, new career choices
have been offered to many young people, especially those who were in less advantaged
positions under the traditional economic structure. Even though it is a very tiny possi-
bility of becoming a millionaire in those fields, it is not hard to earn a relatively satisfac-
tory income through the new economy as a business owner, a video blogger, or an online
streamer.
1
China Broadband Development Alliance (2013), China Internet Speed Condition Report(No.1).
2
China Broadband Development Alliance(2019), China Internet Speed Condition Report(No.24).
69
I will exploit the impacts of Internet speed increase on self-employment choice and
self-employment income using Internet speed data at the province level from the Chinese
Broadband Development Alliance and individual survey data from CFPS. Because of the
data restriction, the variation of internet speed is only at the province level, but the varia-
tion within the individual level is rich in both cross-sectional and over time. Hence I will
use DID (Difference in Differences) specifications to identify those impacts.
This specification requires that individuals in higher-Internet-speed provinces do not
more likely to be self-employed or in different trends. To test that there were no pre-trend
on self-employment before the dramatic increase in internet speed, which could lead to
an inverse causality problem, I did a pre-trend check of Internet speed over province level
on self-employment ratio. Results show that the Internet speed increase is not statistically
significantly related to the self-employment ratio in total employment.
Results are mixed, there are positive effects of internet speed increase on family self-
employment choice and business income in regression without year and province fixed
effects, but these coefficients turn out to be negative and insignificant after including those
fixed effects.
As for individual self-employment choice, increased internet speed is helping individ-
uals to establish entrepreneurship, and the effects are larger for females, for less educated
people, and for people living in rural areas.
The rest of the paper is organized as follows, the second part is background and data
description; the third part is empirical strategy; the fourth part is regression results; and
the last part is discussion.
3.2 Background and Data
Building network facilities have been a main task of the Chinese government within the
past decade, so even though started relatively late compared with South Korea, Japan, and
70
other developed countries on network construction, China has caught up with them re-
cently. The weighted average network download speed increases more than 10 times from
2013 to 2019. Concerning Mobile Network, China has built the largest Forth-Generation
(4G) Mobile Network in the world, and stay in a leading position in the competition with
the Fifth-Generation (5G) Mobile Network.
The introduction of a fast network breeds new online businesses like online streamers,
easy-to-set-up small online businesses, and short video sharing. The development of the
Internet has decentralized normal business structure and each individual can easily start
their own small business with very little initial capital, especially for young low-educated
people who were in disadvantaged business status before. With the help of low-cost fixed
or mobile networks and this new business platform, people have the choice to be self-
employed, instead of low-wage manufacturing jobs.
In an official report released by Cyberspace Administration of China
3
, by June 2019,
the internet penetration rate in China was 61.2%, and more than 99% of them are using
cellphone network. More than 225 million people in rural areas are also internet users.
As for the new industries, 639 million people are shopping online and almost all of them
are doing so on their phones. 633 million people are using mobile payment with phones,
and more than 421 million people are using online food delivery services. Moreover, 648
million people are watching online short videos like TikTok. The Internet development has
been changing China into a huge market for all new industries, and also providing great
opportunities for disadvantaged people to increase their income by establishing their own
entrepreneurship.
I obtained network speed data from the Chinese Broadband Development Alliance, a
quasi-official research institute that regularly collects information on internet speed and
penetration. This institute has been publishing the ”National Report on Broadband Net-
work Speed” quarterly since 2013. Internet speed is reported at the provincial level, where
3
Cyberspace Administration of China(2019), China Internet Development Condition Statistic Report(No.44),
http://www.cac.gov.cn/pdf/20190829/44.pdf.
71
the average internet speed in each province is a population-weighted average of download
or upload speeds across all prefectures. At the prefectural level, the average internet speed
is calculated as a simple average across all time periods and users within the same prefec-
ture, from both urban and rural areas. Since there are more Internet users in cities than in
rural areas, this Internet speed information is biased and could overestimate the network
speed in rural areas.
The Chinese government’s Internet Development Condition Statistic Report also sources its
internet speed data from the Chinese Broadband Development Alliance’s reports, attest-
ing to the credibility of this data. The dataset primarily contains quarterly fixed network
download speed data for each Chinese province from 2012 to 2018, aligning with the four
waves of the China Family Panel Survey (CFPS) data. Mobile network speeds are avail-
able starting from the third quarter of 2016, but I mainly use fixed network speed data,
even though mobile network speed could have been a more suitable index. Given that the
individual-level data represents annual income conditions, I calculate the annual average
internet speed using data from all four quarters. Due to data limitations, the earliest in-
ternet speed data is from the beginning of 2013, so I use this to represent the year 2012.
However, if regression results are significant, using the 2013 internet speed data for 2012
should not be problematic, as it would underestimate the impact of internet speed, if any.
The individual self-employment data is sourced from the China Family Panel Survey
(CFPS), a large-scale panel survey dataset that covers over 57,000 individuals, represent-
ing more than 95% of China’s total population. The survey was first conducted in 2010 and
has been updated every other year, comprising six waves of data. In this paper, I utilize
survey data from 2012, 2014, 2016, and 2018. The dataset includes approximately 16,000
households from 25 provinces, both urban and rural, with rich demographic information
that is either constant or varies over time. The CFPS dataset consists of four data files:
adults, children, family economic condition, and cross-family information. It provides in-
formation on whether an individual is self-employed as a primary job or a main part-time
72
job, as well as details on family entrepreneurship. However, self-employment income is
only surveyed at the family level, and individual-level income data is unavailable. To ad-
dress this, I analyze self-employment career choices at the individual level, considering
heterogeneity, and combine family information with internet speed data to explore the
effects of increasing internet speeds on self-employed income and assets.
Average Value
Urban
0.47
(0.50)
Family Size
3.87
(1.86)
Family Owning House(s)
0.89
(0.31)
Agricultural Work
0.54
(0.50)
Self-employment Indicator
0.09
(0.29)
Self-employment Profit
3708.83
(30801.4)
Self-employment Income
8040.78
(33103.39)
Self-employment Asset
(*10000)
13.07
(2303.16)
Family Total Wage
34567.49
(73765.1)
Family Asset Income
1395.75
(9507.75)
Family Cash and Saving
38803.63
(136169.6)
Observation 47544
Table 3.1: Summary Statistics–Households
Note: All family information comes from four waves of survey data of CFPS, including surveys conducted in the year 2012, 2014, 2016,
and 2018. Standard errors are in parentheses. The family owning a house equals 1 if the family owns the house they are living in fully
or partially. The self-employment indicator equals 1 if the family operates at least one self-employment business. Standard errors are
in parentheses.
73
Average Value
Gender
0.50
(0.50)
Urban
0.47
(0.50)
Age
47.01
(16.50)
Marriage
0.81
(0.39)
Self-employment
0.11
(0.31)
High School Graduate
0.24
(0.43)
School Year
7.03
(4.85)
Spouse High School Graduate
0.53
(0.50)
Agricultural Work
0.30
(0.46)
Using Cell-phone
0.87
(0.33)
Using Mobile Network
0.44
(0.50)
Using Computer Network
0.21
(0.41)
Using Internet
0.39
(0.49)
Observations 131,996
Table 3.2: Summary Statistics–Individuals
Note: All adult information comes from four waves of survey data of China Family Panel Survey (CFPS), including surveys
conducted in the year 2012, 2014, 2016, and 2018. Standard errors are in parentheses.
74
In table 3.1 and table 3.2, descriptive statistics of both family and adult information
are listed. 47% of the families in this survey are from urban areas, and the average family
size is 3.87. 89% of the families are living in a house they own themselves whole or par-
tially. 54% of the families are doing agricultural work, which is more than the portion of
urban families, and this is because some families are living in small towns but still doing
some agricultural work. About 9% of these families own some kind of self-employment
business, with an average income of 8340.78 Yuan, a profit of 3708.83 Yuan, and an av-
erage asset of 130,700 Yuan. Besides these, the average family wage income is 34567.49
Yuan, and the average family investment income is 1395.75 Yuan, which is relatively low
compared with other sources of income. The average cash or saving are 38803.63 Yuan.
As for individual information, I only include adult individuals not still studying at
school at the survey time. 50% of them are male, and 47% are from urban areas. The
average age is 47.01, and 81% of them are married, which only includes active marriage
where their spouse is alive. 11% of these adults are self-employed, including their main
part-time job as self-employed. The average educational status is low, where about 24%
graduated from at least high school, with an average of 7.03 school years. However, 53%
of their spouses are high school or higher graduates. 30% of individuals are doing agricul-
tural work. As for internet usage, 87% of the individuals are using cellphones, consistent
with the population statistics, and 44% of them are mobile network users, about half of
cellphone users. The lower part of the total individuals is using computer networks, 21%.
In the survey of 2012, there is only a merged question of internet usage, and I created a
similar index for other waves, using either mobile or computer networks, and the average
value is 39%. Furthermore, 32% of them are using the Internet to work at least once a
week, 31% to study, 65% to do social interaction, 62% to entertain, and 21% to do business,
all at least once a week. About 33% of them are using emails to communicate.
75
To check the pre-trend of self-employment at the province level, I test whether the
increase in Internet speed is associated with the province self-employment ratio. The ad-
ministrative data at the province level are from the China Data website operated by the
National Bureau of Statistics of China
4
.
3.3 Empirical Strategy and identification challenge
The main empirical specification there is a difference in difference(DID) model. There
could likely be provincial-specific heterogeneity that affects both internet speed and indi-
vidual self-employment choice and income, for example, richer areas have higher budgets
to build network facilities and the market for small businesses is also larger. Inclusion of
provincial and time fixed effects can control these effects and I can capture the within-
province relation between network speed and self-employment income. There are some
other assumptions of identification in this specification. At first, different provinces should
have the same time trends before and after the main index change, otherwise, we could
not attribute the change in self-employment income to network speed change. Another
assumption is that there are no structural changes in the economy across different peri-
ods.
The baseline empirical model will be the following:
Y
ipt
=β
0
+β
1
S
pt
+β
2
S
pt
X
ipt
+ ΓX
ipt
+δ
p
+δ
t
+ϵ
ipt
(3.1)
WhereY
ipt
is the outcome variables including self-employment status and self-employment
income of individual or familyi in provincep at yeart. S
pt
is provincial network speed
of provincep at yeart. X
ipt
is individuali’s controls. δ
p
is province fixed effect. δ
t
is year
fixed effect, and ϵ
ipt
is error terms.
4
http://data.stats.gov.cn/.
76
β
1
will be the main parameter I am focusing on here, which represents the causal re-
lationship between internet speed and self-employment status or income. The third term
is the effect of individual characteristics interacting with network speed, which can show
heterogeneous effects of internet speed on self-employment. I will focus on the effects of
network speed on self-employment conditions for different genders and educational lev-
els. The controls are basic variables that could affect individual career choice, like gender,
education, marriage status, income from other sources, whether doing agricultural work
et al.
The main identification challenge here is that different provinces could have different
pre-trends before the internet speed changes. Then richer provinces have higher network
speed increases and also have residents with higher self-employment income, but it is pos-
sible that there was no causal relation between them. These individuals have higher self-
employment income only because these provinces were in higher pre-trends before the
internet speed increase. To check this problem, I did a regression over the self-employed
ratio in different provinces on internet speed to see if different provinces have different
trends.
Outcome Average Internet Speed (MB/s) Busy Time Internet Speed (MB/s) Free Time Internet Speed (MB/s)
OLS Fixed Effects OLS Fixed Effects OLS Fixed Effects
Self-employed Ratio
0.251
(0.08)
-0.096
(0.652)
0.335
(0.186)
0.243
(0.08)
-0.106
(0.062)
0.318
(0.182)
0.263
(0.087)
-0.082
(0.071)
0.377
(0.197)
GDP per capita
-0.00017
(0.000015)
0.000000
(0.000016)
-0.00018
(0.000028)
-0.00016
(0.000015)
0.0000000
(0.0000148)
-0.00017
(0.000027)
-0.00017
(0.000016)
0.000000
(0.000017)
-0.00019
(0.00003)
Disposable Income
0.00051
(0.000043)
0.000067
(0.000057)
0.00708
(0.000092)
0.000496
(0.000042)
0.00006
(0.0000546)
0.00068
(0.000089)
0.00053
(0.000045)
0.000045
(0.000063)
0.00073
(0.000097)
Time FE No Yes No No Yes No No Yes No
Province FE No No Yes No No Yes No No Yes
Observations 93 93 93 93 93 93 93 93 93
Mean of Outcomes 5.70 5.70 5.70 5.53 5.53 5.53 6.09 6.09 6.09
Table 3.3: Pre-trend Check in Province Level
Note: All family information comes from four waves of survey data of CFPS, including surveys conducted in the year 2012, 2014, 2016,
and 2018. Standard errors are in parentheses. The family owning a house equals 1 if the family owns the house they are living in fully
or partially. The self-employment indicator equals 1 if the family operates at least one self-employment business. Standard errors are
in parentheses.
We can see the regression results in Table 3, the coefficients of the self-employed ratio
are all significant for three outcomes in regression in pooled regression. However, when
77
including either year or province fixed effects, all coefficients lost their significance, in a
confidence level of 5%. I showed that internet speed is not significantly related to the self-
employment ratio in total employment or provincial economic conditions. This shows that
the main assumption of the DID specification in this paper is feasible.
3.4 Main Results
3.4.1 Impacts on Family Self-employment Choice
In the dataset of CFPS, there are family economic conditions surveys, including the ques-
tion of whether the family is running a business or not. Combining this information with
internet speed data and family characteristics, I can detect the impacts of internet speed
increase on self-employment choice, and any heterogeneity of these effects, if any.
Outcome Owner of business
OLS Probit Logit OLS Probit Logit OLS Logit
Average Internet Speed
0.00022
(0.00017)
0.0012
(0.0009)
0.002
(0.002)
0.00004
(0.0002)
0.00004
(0.001)
-0.0002
(0.002)
-0.0008
(0.0019)
0.009
(0.045)
Busy Internet Speed
0.00022
(0.00017)
Non-busy Internet Speed
0.00023
(0.00016)
Urban
0.007
(0.002)
0.049
(0.011)
0.11
(0.03)
-0.002
(0.002)
-0.043
(0.036)
Family Size
0.025
(0.0008)
0.14
(0.005)
0.026
(0.009)
0.014
(0.001)
0.27
(0.03)
Agricultural Work
-0.054
(0.003)
-0.33
(0.02)
-0.63
(0.04)
-0.019
(0.004)
-0.36
(0.10)
Family Owning House(s)
-0.012
(0.004)
-0.07
(0.03)
-0.14
(0.05)
-0.006
(0.005)
-0.16
(0.11)
Total Family Wage Income
-0.007
(0.0003)
-0.04
(0.002)
-0.08
(0.003)
-0.004
(0.0003)
-0.07
(0.01)
Total Family Investment Income
0.006
(0.0004)
0.03
(0.003)
0.06
(0.005)
0.002
(0.0006)
0.03
(0.01)
Total Family Cash and Saving
0.004
(0.0003)
0.03
(0.002)
0.05
(0.004)
0.002
(0.0003)
0.05
(0.01)
Province FE No No No No No No No No Yes Yes
Year FE No No No No No No No No Yes Yes
Observation 47544 47544 47544 47544 47544 44936 44936 44936 44936 6525
Mean of Outcome 0.094 0.094 0.094 0.094 0.094 0.094 0.094 0.094 0.094 0.094
Table 3.4: Internet Speed and Family Self-Employment Choice
Note: All family information comes from four waves of survey data of CFPS, including surveys conducted in the year 2012, 2014, 2016,
and 2018. Standard errors are in parentheses. The family owning a house equals 1 if the family owns the house they are living in fully
or partially. The self-employment indicator equals 1 if the family operates at least one self-employment business. Standard errors are
in parentheses.
78
In table 3.4, there are results of regression on whether a family is owning a business
over internet speed and other controls. We can see from the first three regressions that
there are nearly no differences in using these three indicators of internet downloading
speed, so I will just show the results of regression where the main independent variable
is the average internet speed. We can see that all coefficients of the internet speed are sta-
tistically significant, at least in a 10% confidence interval in the first three columns. When
including all controls, and no fixed effects, the coefficients are no longer significant. When
including controls and both province-fixed effects and time-fixed effects, the coefficients
of average internet speed are still insignificant.
As for control variables included in the regression here, all of them are strongly signif-
icant when fixed effects are not included. The coefficients of urban, family size and family
investment income are positive. It shows urban families, families with more members and
higher investment income are more likely to own a business. In contrast, the coefficients
of doing agricultural work, owning houses, and family total wage income are negative.
These show that families doing agricultural work and more wage work are less likely to
have self-employed businesses. Also, families owning their own houses are less likely to
have a business, which is not what I expected. The significant level for some variables
turns out to be 10% when including year and province fixed effects. The effect of urban
turns negative and less significant.
3.4.2 Impacts on Self-employment Income and Assets
Then using family-level information in self-employed income and business assets varia-
tion, I can explore the impacts of internet speed increase on family business income and
business assets.
In table 3.5, I show the results of regression on family self-employed income and assets
over average internet speed to see the effects of internet downloading speed on family self-
employed income and related asset amounts. First, for self-employed income, in pooled
79
Outcome Self-employment Income Self-employment Assets
Pooled FE Pooled FE Pooled FE Pooled FE
Average Internet Speed
-0.05
(0.003)
0.03
(0.03)
-0.04
(0.002)
0.046
(0.026)
0.003
(0.002)
-0.003
(0.19)
0.001
(0.002)
0.004
(0.02)
Urban
-0.005
(0.02)
0.007
(0.024)
0.08
(0.018)
-0.025
(0.018)
Family Size
0.22
(0.01)
0.11
(0.02)
0.26
(0.01)
0.14
(0.014)
Agricultural Work
5.43
(0.04)
4.95
(0.06)
-0.57
(0.03)
-0.22
(0.046)
Family Owning House(s)
0.06
(0.05)
0.13
(0.07)
-0.17
(0.05)
-0.06
(0.05)
Total Family Wage Income
-0.06
(0.004)
-0.02
(0.005)
-0.07
(0.003)
-0.04
(0.004)
Total Family Investment Income
0.03
(0.006)
-0.007
(0.008)
0.06
(0.01)
0.02
(0.01)
Total Family Cash and Saving
0.08
(0.004)
0.06
(0.07)
0.05
(0.003)
0.03
(0.003)
Province FE No Yes No Yes No Yes No Yes
Year FE No Yes No Yes No Yes No Yes
Observation 46506 46506 44073 44073 47218 47218 44719 44719
Mean of Outcome 8040.98 8040.98 8040.98 8040.98 130684.2 130684.2 130684.2 130684.2
Table 3.5: Internet Speed and Family Self-Employment Income and Assets
Note: All family information comes from four waves of survey data of CFPS, including surveys conducted in the year 2012, 2014, 2016,
and 2018. Standard errors are in parentheses. The family owning a house equals 1 if the family owns the house they are living in fully
or partially. The self-employment indicator equals 1 if the family operates at least one self-employment business. Standard errors are
in parentheses.
regression and regression including fixed effects, the coefficients are insignificant. How-
ever, in the regression with all control variables and year and province fixed effects, the
result shows that internet speed increase is positively correlated with family self-employed
income, at least at a 10% confidence level. For significant other controls, families with big-
ger size, owning houses, and doing agricultural work have higher self-employed income.
The coefficient of doing agricultural work turns out to be positive here, which could show
that even though families doing agricultural work are less likely to own their own busi-
ness, they can have a high self-employed income if they choose to own a business. The
result for the family’s total cash or saving is positive, even though insignificant.
As for the regression of self-employed business assets, the results of the first three
columns are still insignificant, and neither for the coefficient of the regression including
all controls and fixed effects. All other controls have the same sign of effects except for agri-
cultural work and owning houses, which are negative for self-employed business assets
80
now. The effects of family cash and saving are significant and positive for self-employed
assets.
In a word, these results show that the increase in internet speed is positively correlated
with family self-employed income, but not assets. This could possibly be because of the
accumulation process of assets, and it could take more time for the increase of income to
lead to assets increase.
3.4.3 Impacts on Individual Self-employment Choice
Then I will show how the increase in internet speed is affecting individual self-employment
choices. In Table 3.6, there are results of regression over individual self-employed status,
as a main job or part-time job, on average internet speed and other individual-level con-
trols.
Outcome Individual Self-employed Choice
OLS Probit Logit OLS Probit Logit OLS Logit OLS
Average Internet Speed
0.0009
(0.0001)
0.005
(0.0006)
0.009
(0.001)
-0.001
(0.001)
-0.007
(0.007)
-0.013
(0.013)
0.0017
(0.0014)
0.031
(0.030)
0.139
(0.053)
Urban
0.030
(0.009)
0.167
(0.052)
0.329
(0.101)
0.119
(0.146)
Age
-0.001
(0.0003)
-0.008
(0.002)
-0.015
(0.004)
-0.146
(0.115)
Gender
0.057
(0.009)
0.336
(0.052)
0.660
(0.103)
-0.328
(0.403)
Marriage Status
-0.016
(0.035)
-0.077
(0.208)
-0.185
(0.399)
0.146
(0.193)
High School Graduate
-0.017
(0.012)
-0.083
(0.064)
-0.176
(0.125)
0.139
(0.375)
Agricultural Work
-0.049
(0.009)
-0.301
(0.056)
-0.582
(0.11)
-0.194
(0.096)
Spouse High School
0.0004
(0.012)
0.001
(0.064)
-0.004
(0.124)
-0.007
(0.202)
Province FE No No No No No No Yes Yes Yes
Year FE No No No No No No Yes Yes Yes
Observation 101,241 101,241 101,241 4,935 4,935 4,935 101,241 14,287 4,935
Mean of Outcome 0.11 0.11 0.11 0.104 0.104 0.104 0.104 0.104 0.104
Table 3.6: Internet Speed and Individual Self-Employment Choice
Note: All adult information comes from four waves of survey data of China Family Panel Survey (CFPS), including surveys
conducted in the year 2012, 2014, 2016, and 2018. Standard errors are in parentheses.
81
The coefficients of average internet speed when excluding all control and fixed effects
are significantly positive. However, when including control variables, the effects of in-
ternet speed are insignificant, and become negative in all three methods. When includ-
ing all controls and two fixed effects, the main result is strongly significant now, at a 5%
confident level. So under increased internet speed, individuals are more likely to choose
self-employment as their main job or part-time job.
As for the controls, only the coefficient of doing agricultural work is significant, and
negative, which is consistent with the result for family choice regression.
In conclusion, the result here shows that internet speed increases can cause individuals
more likely to be self-employed, or at least choose self-employment as a main part-time
job if it’s not their main job.
3.4.4 Heterogeneity
In this part, I will show the heterogeneity of the effects of an internet speed increase over
different individuals, including gender, educational level, age, marriage status, urban in-
dex, agricultural work index, and spouse educational level.
In Table 3.7, we can see that the coefficients of average internet speed stay significantly
positive in all specifications. However, the coefficients of intersections are all insignificant.
The coefficients for urban index, age, marriage, gender, and doing agricultural work are
negative. This implies that female adults, individuals from rural areas, younger individu-
als, unmarried individuals and individuals not doing agricultural work can benefit more
from the increase in internet speed, in terms of self-employed choice. The coefficients of
high school graduate and spouse of high school graduate are positive, which imply that in-
dividuals with better educational status can benefit more from the internet speed increase.
However, none of the implications above are proven with significant results.
82
Outcomes
Individual Self-employed
OLS
Average Internet Speed
0.133
(0.053)
0.148
(0.059)
0.160
(0.059)
0.156
(0.065)
0.149
(0.055)
0.145
(0.054)
0.141
(0.054)
Average Internet Speed
x Urban
-0.011
(0.009)
Average Internet Speed
x Age
-0.0001
(0.0004)
Average Internet Speed
x Gender
-0.007
(0.009)
Average Internet Speed
x Marriage
-0.014
(0.029)
Average Internet Speed
x High School Graduate
0.008
(0.011)
Average Internet Speed
x Agricultural Work
-0.131
(0.121)
Average Internet Speed
x Spouse High School
0.010
(0.012)
Urban
0.208
(0.162)
0.130
(0.059)
0.142
(0.149)
0.117
(0.147)
0.114
(0.147)
0.138
(0.148)
0.103
(0.148)
Age
-0.131
(0.115)
-0.150
(0.115)
-0.148
(0.115)
-0.131
(0.120)
-0.152
(0.115)
-0.148
(0.115)
-0.132
(0.116)
Gender
-0.281
(0.403)
-0.327
(0.407)
-0.303
(0.406)
-0.323
(0.407)
-0.291
(0.408)
-0.363
(0.406)
-0.265
(0.412)
Marriage
0.131
(0.192)
0.147
(0.194)
0.165
(0.195)
0.302
(0.377)
0.133
(0.194)
0.114
(0.197)
0.101
(0.201)
High School Graduate
-0.010
(0.392)
0.166
(0.386)
0.165
(0.378)
0.173
(0.384)
0.008
(0.011)
0.132
(0.376)
0.001
(0.411)
Agricultural Work
-0.165
(0.098)
-0.203
(0.101)
-0.207
(0.098)
-0.202
(0.098)
-0.196
(0.097)
-0.131
(0.121)
-0.203
(0.097)
Spouse High School
0.028
(0.203)
0.012
(0.211)
-0.018
(0.203)
-0.015
(0.204)
0.018
(0.206)
0.037
(0.209)
-0.081
(0.222)
Year FE Yes Yes Yes Yes Yes Yes Yes
Province FE Yes Yes Yes Yes Yes Yes Yes
Oberservations 4,935 4,935 4,935 4,935 4,935 4,935 4,935
Mean of Outcome 0.11 0.11 0.11 0.11 0.11 0.11 0.11
Table 3.7: Heterogeneity in Impacts of Individual Self-Employment Choice
Note: All adult information comes from four waves of survey data of China Family Panel Survey (CFPS), including surveys
conducted in the year 2012, 2014, 2016, and 2018. Standard errors are in parentheses.
83
3.5 Discussion
Compared with traditional wage jobs, self-employment could be a preferred alternative
choice for low-income people in developing areas, specifically for less advantaged groups,
like less educated females living in rural areas.
However, there are lots of restrictions for establishing profitable entrepreneurship in
those poor areas like credit constraints, market restrictions, skill shortages, and cultural
restrictions. Technological development could be a potential solution to some restrictions
those entrepreneurs are facing. Popularization of faster internet connection is making es-
tablishing entrepreneurship much easier nowadays because of the easier process of build-
ing a business, easier access to customers, and fewer professional skills requirement.
Using quarterly internet speed data from China Broadband Development Alliance and
individual and family level information from the large-scale panel survey data from CFPS,
I explore the impacts of increasing internet speed on both self-employment choices and
family self-employment income and assets.
We can not see significant coefficients for self-employment choice in the family re-
gression, however, results show that individuals are more likely to be self-employed with
higher internet speed. As for families owning businesses, we can see they have more in-
come with faster internet speed, but not higher business assets.
As for the heterogeneity check, none of the results are significant. The coefficients of
these intersection parameters of the urban index, age, marriage status, and doing agricul-
tural work are negative, while the coefficients of educational status for both themselves
and the spouses’ are positive. These can be interpreted as the effects for younger individ-
uals, females, individuals from rural areas, unmarried individuals, and higher educated
individuals are larger, even though it should be noted that none of those results are sig-
nificant.
84
Conclusion
In conclusion, this dissertation has examined various facets of development economics in
China, with a focus on the intersections between inequality and the labor market. It cov-
ers the impacts of fertility on female labor market outcomes, using the national population
policy change as an exogenous shock; the trends in education returns and income inequal-
ity, and the relationship between technological advancements and self-employment deci-
sions.
The first chapter demonstrated the unintended consequences of China’s one-child pol-
icy relaxation on women’s labor market outcomes, emphasizing the importance of con-
sidering fertility expectations and their broader implications when designing population
policies. The analysis revealed that prime-age women and policy compliers experienced
the most pronounced declines in labor participation, work hours, and salary earnings,
with additional factors like education, community types, occupations, and family income
exacerbating the impacts. Results are robust using the recent heterogeneity robust estima-
tion methods and show that the national population policy change could have unintended
broader impacts on females, and also on equity. Future research will continue to provide
more direct evidence on labor market discrimination against affected females, and also
further implications on the marriage market and migration decisions.
The second chapter explored income inequality and the heterogeneous returns to ed-
ucation in China, showcasing a 54% decline in overall education return in the past decade
and increased dispersion across various divides. These findings suggest potential future
divergence in wages and household incomes, emphasizing the need for policymakers
85
to address educational and income disparities in both rural and urban areas, as well as
among different educational attainment levels and gender groups. Future research will
continue to do further decomposition of income inequalities, and to explore the implica-
tions of declining education returns, especially on education investment.
Finally, the third chapter investigated the correlation between high-speed internet ac-
cess and self-employment choices in China. The results indicated that increased inter-
net speed is associated with a higher likelihood of self-employment, particularly among
women, lower-educated individuals, and rural residents, highlighting the potential of
technology to create alternative income opportunities for disadvantaged groups. How-
ever, no evidence was found linking Internet speed with business assets or income, indi-
cating that further research is necessary to understand the nuances of this relationship, as
well as detailed mechanisms behind the role of Internet speed.
In summary, this dissertation has contributed to the understanding of development
economics in China by shedding light on the interconnectedness of fertility, education,
income inequality, and technology. The findings highlight the need for more inclusive and
targeted policy interventions that address both gender and income disparities in order to
foster a more equitable and sustainable development trajectory for China.
86
References
Goldin, Claudia (2006). “The Quiet Revolution That Transformed Women’s Employment,
Education, and Family”. In: American Economic Review 96 (2), pp. 1–21.
Blau, Francine D. and Lawrence M. Kahn (2017). “The Gender Wage Gap: Extent, Trends,
and Explanations”. In: Journal of Economic Literature 55 (3), pp. 789–865.doi:10.1257/
JEL.20160995.
Kleven, Henrik and Camille Landais (2017). “Gender Inequality and Economic Devel-
opment: Fertility, Education and Norms”. In: Economica 84 (334), pp. 180–209. doi:
10.1111/ECCA.12230.
Kleven, Henrik, Camille Landais, Johanna Posch, Andreas Steinhauer, and Josef Zweim¨ uller
(2019). “Child Penalties across Countries: Evidence and Explanations”. In: AEA Papers
and Proceedings 109, pp. 122–26.doi:10.1257/PANDP.20191078.
Cort´ es, Patricia and Jessica Pan (2020). Children and the remaining gender gaps in the labor
market. Tech. rep. National Bureau of Economic Research.
Rosenzweig, Mark R and Kenneth I Wolpin (1980). “Life-cycle labor supply and fertil-
ity: Causal inferences from household models”. In: Journal of Political Economy 88.2,
pp. 328–348.
87
Angrist, Joshua D and William N Evans (1998). “Children and Their Parents’ Labor Sup-
ply: Evidence from Exogenous Variation in Family Size”. In: The American Economics
Review 88 (3), pp. 450–477.
Lundborg, Petter, Erik Plug, and Astrid W¨ urtz Rasmussen (2017). “Can Women Have
Children and a Career? IV Evidence from IVF Treatments”. In: American Economic Re-
view 107 (6), pp. 1611–1637.doi:10.1257/aer.20141467.
Markussen, Simen and Marte Strøm (2022). “Children and labor market outcomes: sepa-
rating the effects of the first three children”. In: Journal of Population Economics 35 (1),
pp. 135–167.doi:10.1007/S00148-020-00807-0/FIGURES/12.
United Nations Department of Economic and Social Affairs, Population Division (2021).
World Population Policies 2021: Policies related to fertility. Tech. rep. UN DESA/POP/2021/TR/NO.
1.
Roth, Jonathan, Pedro H C Sant’anna, Alyssa Bilinski, and John Poe (2022). “What’s Trend-
ing in Difference-in-Differences? A Synthesis of the Recent Econometrics Literature”.
In: Working Paper.
Chaisemartin, Cl´ ement de and Xavier D’haultfoeuille (2022). “Two-Way Fixed Effects and
Differences-in-Differences with Heterogeneous Treatment Effects: A Survey”. In: Work-
ing Paper.
Goodman-Bacon, Andrew (2021). “Difference-in-differences with variation in treatment
timing”. In: Journal of Econometrics.doi:10.1016/j.jeconom.2021.03.014.
Chaisemartin, Cl´ ement de and Xavier D’Haultfœuille (2020). “Two-Way Fixed Effects Es-
timators with Heterogeneous Treatment Effects”. In: American Economic Review 110 (9),
pp. 2964–2996.doi:10.1257/aer.20181169.
88
Callaway, Brantly and Pedro H.C. Sant’Anna (2021). “Difference-in-Differences with mul-
tiple time periods”. In: Journal of Econometrics 225 (2), pp. 200–230. doi: 10.1016/j.
jeconom.2020.12.001.
Altonji, Joseph G. and Rebecca M. Blank (1999). Race and gender in the labor market. Vol. 3.
Elsevier, pp. 3143–3259.doi:10.1016/S1573-4463(99)30039-0.
Bertrand, Marianne (2011). New Perspectives on Gender. Vol. 4. Elsevier, pp. 1543–1590.doi:
10.1016/S0169-7218(11)02415-4.
Olivetti, Claudia and Barbara Petrongolo (2016). “The Evolution of Gender Gaps in In-
dustrialized Countries”. In: Annual Review of Economics 8, pp. 405–434. doi: 10.1146/
ANNUREV-ECONOMICS-080614-115329.
Bertrand, Marianne (2020). “Gender in the Twenty-First Century”. In: AEA Papers and Pro-
ceedings, pp. 1–24.doi:10.1257/pandp.20201126.
Angelov, Nikolay, Per Johansson, and Erica Lindahl (2016). “Parenthood and the gender
gap in pay”. In: Journal of Labor Economics 34 (3), pp. 545–579. doi: 10.1086/684851/
SUPPL_FILE/13148DATA.ZIP.
Kleven, Henrik, Camille Landais, and Jakob Egholt Søgaard (2019). “Children and Gender
Inequality: Evidence from Denmark”. In: American Economic Journal: Applied Economics
11 (4), pp. 181–209.doi:10.1257/APP.20180010.
Kim, Scott Daewon and Petra Moser (2021). “Women in Science. Lessons From the Baby
Boom”. In: NBER Working Paper No. 29436.
Berniell, In´ es, Lucila Berniell, Dolores de la Mata, Mar´ ıa Edo, and Mariana Marchionni
(2021). “Gender gaps in labor informality: The motherhood effect”. In: Journal of De-
velopment Economics 150, p. 102599.doi:10.1016/J.JDEVECO.2020.102599.
89
Meekes, Jordy and Wolter H.J. Hassink (2022). “Gender differences in job flexibility: Com-
mutes and working hours after job loss”. In: Journal of Urban Economics 129, p. 103425.
doi:10.1016/J.JUE.2022.103425.
Cort´ es, Patricia and Jessica Pan (2019). “When time binds: Substitutes for household pro-
duction, returns to working long hours, and the skilled gender wage gap”. In: Journal of
Labor Economics 37 (2), pp. 351–398.doi:10.1086/700185/SUPPL_FILE/16287DATA.ZIP.
Akyol, Pelin and Zeynep Yilmaz (2021). “Effects of Grandmothers’ Proximity on Mothers’
Labor Force Participation”. In: Working Paper.
Talamas, Miguel (2022). “Grandmothers and the Gender Gap in the Mexican Labor Mar-
ket”. In: IDB Working Paper Series No. 1337.
Kleven, Henrik, Camille Landais, and Jakob Egholt Søgaard (2021). “Does Biology Drive
Child Penalties? Evidence from Biological and Adoptive Families”. In: American Eco-
nomic Review: Insights 3 (2), pp. 183–98.doi:10.1257/AERI.20200260.
Andresen, Martin Eckhoff and Emily Nix (2022). “What Causes the Child Penalty? Evi-
dence from Adopting and Same-Sex Couples”. In: Journal of Labor Economics 40 (4).doi:
10.1086/718565.
Levendis, John and Aaron Lowen (2022). “What Same-Sex Adoption Laws Can Tell Us
About the Gender Wage Gap in the United States”. In: Journal of Family and Economic
Issues 1, pp. 1–17.doi:10.1007/s10834-022-09835-2.
Ginja, Rita, Arizo Karimi, and Pengpeng Xiao (2023). “Employer responses to family leave
programs”. In: American Economic Journal: Applied Economics 15.1, pp. 107–35.
Brenøe, Anne A, Serena Canaan, Nikolaj A Harmon, and Heather N Royer (2020). “Is
Parental Leave Costly for Firms and Coworkers?” In: NBER Working Paper No. 26622.
90
Zhang, Junsen (2017). “The evolution of China’s one-child policy and its effects on family
outcomes”. In: Journal of Economic Perspectives 31 (1), pp. 141–160. doi: 10.1257/jep.
31.1.141.
Li, Hongbin, Junsen Zhang, and Yi Zhu (2005). “The Effect of the One-Child Policy on
Fertility in China: Identification Based on the Differences-in-Differences”. In: Working
Paper.
Cai, Yong (2010). “China’s Below-Replacement Fertility: Government Policy or Socioeco-
nomic Development?” In: Population and Development Review 36 (3), pp. 419–440. doi:
10.1086/681664.
Li, Hongbin, Junsen Zhang, and Yi Zhu (2008). “THE QUANTITY-QUALITY TRADE-
OFF OF CHILDREN IN A DEVELOPING COUNTRY: IDENTIFICATION USING CHI-
NESE TWINS”. In: Demography 45 (1), pp. 223–243.
Rosenzweig, Mark R. and Junsen Zhang (2009). “Do population control policies induce
more human capital investment? Twins, birth weight and China’s ”one-child” policy”.
In: Review of Economic Studies 76 (3), pp. 1149–1174.doi:10.1111/j.1467-937X.2009.
00563.x.
Li, Bingjing and Hongliang Zhang (2017). “Does population control lead to better child
quality? Evidence from China’s one-child policy enforcement”. In: Journal of Compara-
tive Economics 45 (2), pp. 246–260.doi:10.1016/J.JCE.2016.09.004.
Huang, Wei, Xiaoyan Lei, and Ang Sun (2021). “Fertility Restrictions and Life Cycle Out-
comes: Evidence from the One-Child Policy in China”. In: The Review of Economics and
Statistics 103 (4), pp. 694–710.doi:10.1162/REST_A_00921.
91
Huang, Wei, Xiaoyan Lei, and Yaohui Zhao (2016). “One-child policy and the rise of man-
made twins”. In: Review of Economics and Statistics 98 (3), pp. 467–476. doi: 10.1162/
REST_a_00567.
Huang, Wei and Yi Zhou (2015). “One-Child Policy, Marriage Distortion, and Welfare
Loss”. In: IZA Discussion Paper Series No. 9532.
Wang, Fei, Liqiu Zhao, and Zhong Zhao (2017). “China’s family planning policies and
their labor market consequences”. In: Journal of Population Economics 30 (1), pp. 31–68.
doi:10.1007/s00148-016-0613-0.
Ebenstein, Avraham (2010). “The ”Missing Girls” of China and the Unintended Conse-
quences of the One Child Policy”. In: The Journal of Human Resources 45 (1), pp. 87–
115.
Li, Hongbin, Junjian Yi, and Junsen Zhang (2011). “Estimating the Effect of the One-Child
Policy on the Sex Ratio Imbalance in China: Identification Based on the Difference-in-
Differences”. In: Demography 48 (4), pp. 1535–1557.doi:10.1007/s13524-011-0055-y.
Qin, Yu and Fei Wang (2017). “Too early or too late: What have we learned from the 30-
year two-child policy experiment in Yicheng, China?” In: Demographic Research 37 (1),
pp. 929–956.doi:10.4054/DemRes.2017.37.30.
He, Haoran, Sherry Xin Li, and Yuling Han (2023). “Labor market discrimination against
family responsibilities: A correspondence study with policy change in China”. In: Jour-
nal of Labor Economics 41.2, pp. 361–387.
Sun, Liyang and Sarah Abraham (2021). “Estimating dynamic treatment effects in event
studies with heterogeneous treatment effects”. In: Journal of Econometrics 225 (2), pp. 175–
199.doi:10.1016/J.JECONOM.2020.09.006.
92
Borusyak, Kirill, Xavier Jaravel, and Jann Spiess (2022). “Revisiting Event Study Designs:
Robust and Efficient Estimation”. In: Working Paper.
Wang, Shaoda and David Yang (2022). “Policy Experimentation in China: The Political
Economy of Policy Learning”. In: Working Paper.
Gan, Li, Qing He, Ruichao Si, and Daichun Yi (2019). “Relocating or Redefined: A New
Perspective on Urbanization in China”. In: NBER Working Paper No. 26585. doi: 10.
3386/W26585.
Ravallion, Martin and Shaohua Chen (2007). “China’s (uneven) progress against poverty”.
In: Journal of Development Economics 82 (1), pp. 1–42. doi: 10.1016/j.jdeveco.2005.
07.003.
Piketty, Thomas and Nancy Qian (2009). “Income Inequality and Progressive Income Tax-
ation in China and India, 1986-2015”. In: American Economic Journal: Applied Economics
1 (2), pp. 53–63.doi:10.1257/APP.1.2.53.
Xie, Yu and Xiang Zhou (2014). “Income inequality in today’s China”. In: Proceedings of
the National Academy of Sciences, pp. 6928–6933.
Knight, John (2014). “Inequality in China: An Overview”. In: The World Bank Research Ob-
server 29 (1), pp. 1–19.doi:10.1093/WBRO/LKT006.
Piketty, Thomas, Li Yang, and Gabriel Zucman (2019). “Capital Accumulation, Private
Property, and Rising Inequality in China, 1978–2015”. In: American Economic Review
109 (7), pp. 2469–96.doi:10.1257/AER.20170973.
Zhang, Junsen (2021). “A Survey on Income Inequality in China”. In: Journal of Economic
Literature 59 (4), pp. 1191–1239.doi:10.1257/JEL.20201495.
93
Chen, Jiandong, Ming Pu, and Wenxuan Hou (2019). “The trend of the Gini coefficient of
China (1978–2010)”. In: Journal of Chinese Economic and Business Studies 17 (3), pp. 261–
285.doi:10.1080/14765284.2019.1663695.
Wan, Guanghua, Ting Wu, and Yan Zhang (2018). “The Decline of Income Inequality in
China: Assessments and Explanations”. In: Asian Economic Papers 17 (3), pp. 115–140.
doi:10.1162/ASEP_A_00640.
Kanbur, Ravi, Yue Wang, and Xiaobo Zhang (2021). “The great Chinese inequality turnaround”.
In: Journal of Comparative Economics 49 (2), pp. 467–482.doi:10.1016/J.JCE.2020.10.
001.
Ravallion, Martin and Shaohua Chen (2022). “Is that really a Kuznets curve? Turning
points for income inequality in China”. In: Journal of Economic Inequality 20 (4), pp. 749–
776.doi:10.1007/S10888-022-09541-X/METRICS.
Kuznets, Simon (1955). “Economic Growth and Income Inequality”. In: The American Eco-
nomic Review 45 (1), pp. 1–28.
Gao, Zhan and M Hashem Pesaran (2023). “Identification and estimation of categorical
random coefficient models”. In: Empirical Economics, pp. 1–46.
Sicular, Terry, Yue Ximing, and Li Shi (2007). “THE URBAN-RURAL INCOME GAP AND
INEQUALITY IN CHINA”. In: Review of Income and Wealth 53 (1), pp. 93–126.doi:10.
1111/j.1475-4991.2007.00219.x.
Liu, Xueliang (2010). “Decomposition of china’s income inequality, 1995-2006”. In: Chinese
Economy 43.4, pp. 49–72.
Lemieux, Thomas (2006a). “Increasing residual wage inequality: Composition effects, noisy
data, or rising demand for skill?” In: The American Economic Review 96.3, pp. 461–498.
94
Lemieux, Thomas (2006b). “Postsecondary education and increasing wage inequality”.
In: American Economic Review 96.2, pp. 195–199.
Heckman, James J, John Eric Humphries, and Gregory Veramendi (2018). “Returns to
Education: The Causal Effects of Education on Earnings, Health, and Smoking”. In:
Journal of Political Economy 126 (S1), S197–S246.
Gustafsson, Bj¨ orn and Li Shi (2002). “Income inequality within and across counties in
rural China 1988 and 1995”. In: Journal of Development Economics 69 (1), pp. 179–204.
doi:10.1016/S0304-3878(02)00058-5.
Benjamin, Dwayne, Loren Brandt, and John Giles (2005). “The Evolution of Income In-
equality in Rural China”. In: Economic development and cultural change 53 (4), pp. 769–
824.
Khan, Azizur Rahman and Carl Riskin (2005). “China’s Household Income and Its Dis-
tribution, 1995 and 2002*”. In: The China Quarterly 182, pp. 356–384. doi: 10.1017/
S0305741005000238.
Sicular, Terry, Yue Ximing, Bj¨ orn Gustafsson, and Li Shi (2007). “The urban–rural income
gap and inequality in China”. In: Review of Income and Wealth 53.1, pp. 93–126.
Cai, Fang (2021). “The Income Distribution Kuznets Turning Point”. In: Understanding
China’s Economy. Springer Books. Springer. Chap. 0, pp. 195–206. doi: 10.1007/978-
981-33-6322-9.
Byron, Raymond P. and Evelyn Q. Manaloto (1990). “Returns to Education in China”. In:
Economic Development and Cultural Change 38 (4), pp. 783–796.doi:10.1086/451833.
Zhao, Yaohui (1997). “Labor Migration and Returns to Rural Education in China”. In:
American Journal of Agricultural Economics 79 (4), pp. 1278–1287.doi:10.2307/1244284.
95
Li, Tianyou and Junsen Zhang (1998). “Returns to education under collective and house-
hold farming in China”. In: Journal of Development Economics 56 (2), pp. 307–335. doi:
10.1016/S0304-3878(98)00068-6.
Zhang, Junsen, Yaohui Zhao, Albert Park, and Xiaoqing Song (2005). “Economic returns
to schooling in urban China, 1988 to 2001”. In: Journal of Comparative Economics 33 (4),
pp. 730–752.doi:10.1016/J.JCE.2005.05.008.
Li, Qiang, Alan De Brauw, Scott Rozelle, and Linxiu Zhang (2005). “Labor Market Emer-
gence and Returns to Education in Rural China”. In: Applied Economic Perspectives and
Policy 27 (3), pp. 418–424.doi:10.1111/J.1467-9353.2005.00238.X.
De Mel, Suresh, David McKenzie, and Christopher Woodruff (2008). “Returns to capi-
tal in microenterprises: evidence from a field experiment”. In: The quarterly journal of
Economics 123.4, pp. 1329–1372.
Li, Hongbin, Pak Wai Liu, and Junsen Zhang (2012). “Estimating returns to education
using twins in urban China”. In: Journal of Development Economics 97 (2), pp. 494–504.
doi:10.1016/J.JDEVECO.2011.05.009.
Li, Haizheng (2003). “Economic transition and returns to education in China”. In: Eco-
nomics of Education Review 22 (3), pp. 317–328.doi:10.1016/S0272-7757(02)00056-0.
Wu, Xiaogang and Yu Xie (2003). “Does the Market Pay off? Earnings Returns to Education
in Urban China”. In: American Sociological Review 68 (3), pp. 425–442.
Liu, Zhiqiang (2007). “The external returns to education: Evidence from Chinese cities”.
In: Journal of Urban Economics 61 (3), pp. 542–564.doi:10.1016/J.JUE.2006.08.007.
96
Fang, Hai, Karen N Eggleston, John A Rizzo, Scott Rozelle, and Richard J Zeckhauser
(2012). “The Returns to Education in China: Evidence from the 1986 Compulsory Ed-
ucation Law”. In: NBER Working Paper Series No. 18189.
Knight, John, Quheng Deng, and Shi Li (2017). “China’s expansion of higher education:
The labour market consequences of a supply shock”. In: China Economic Review 43,
pp. 127–141.doi:10.1016/J.CHIECO.2017.01.008.
Asadullah, M Niaz and Saizi Xiao (2019). “Labor Market Returns to Education and En-
glish Language Skills in the People’s Republic of China: An Update”. In: Asian Devel-
opment Review 36 (1), pp. 80–111.
Hu, Anning and Jacob Hibel (2014). “Changes in college attainment and the economic
returns to a college degree in urban China, 2003–2010: Implications for social equality”.
In: Social Science Research 44, pp. 173–186.doi:10.1016/J.SSRESEARCH.2013.12.001.
Rosenzweig, Mark R. and Junsen Zhang (2013). “Economic growth, comparative advan-
tage, and gender differences in schooling outcomes: Evidence from the birthweight
differences of Chinese twins”. In: Journal of Development Economics 104, pp. 245–260.
doi:10.1016/J.JDEVECO.2013.01.002.
Messinis, George (2013). “Returns to education and urban-migrant wage differentials in
China: IV quantile treatment effects”. In: China Economic Review 26 (1), pp. 39–55.doi:
10.1016/J.CHIECO.2013.03.006.
Wang, Xiaojun, Belton M. Fleisher, Haizheng Li, and Shi Li (2014). “Access to college and
heterogeneous returns to education in China”. In: Economics of Education Review 42,
pp. 78–92.doi:10.1016/J.ECONEDUREV.2014.05.006.
Gao, Wenshu and Russell Smyth (2014). “Education expansion and returns to schooling in
urban China, 2001–2010: evidence from three waves of the China Urban Labor Survey”.
97
In: Journal of the Asia Pacific Economy 20 (2), pp. 178–201.doi:10.1080/13547860.2014.
970607.
Sch¨ oEmann, Klaus and Rolf Becker (2015). “Returns to Education in Different Labor Mar-
ket Contexts”. In: Emerging Trends in the Social and Behavioral Sciences, pp. 1–13. doi:
10.1002/9781118900772.ETRDS0290.
Pi, Jiancai and Pengqing Zhang (2018). “Rural–urban human capital disparity and skilled–unskilled
wage inequality in China”. In: Review of Development Economics 22 (2), pp. 827–843.doi:
10.1111/RODE.12370.
Liu, Lin, Krishna P. Paudel, Guanghao Li, and Ming Lei (2019). “Income inequality among
minority farmers in China: Does social capital have a role?” In: Review of Development
Economics 23 (1), pp. 528–551.doi:10.1111/RODE.12559.
Churchill, Sefa Awaworyi and Vinod Mishra (2018). “Returns to education in China: a
meta-analysis”. In: https://doi.org/10.1080/00036846.2018.1488074 50 (54), pp. 5903–
5919.doi:10.1080/00036846.2018.1488074.
Ma, Xinxin and Ichiro Iwasaki (2021). “Return to schooling in China: a large meta-analysis”.
In: Education Economics 29 (4), pp. 379–410. doi: 10.1080/09645292.2021.1900791/
SUPPL_FILE/CEDE_A_1900791_SM3360.XLSX.
Chen, Shuxing, Lingfeng Zheng, and Yuxiang Gao (2022). “Educational divergence among
urban residents and migrant workers: Evidence from China”. In: Chinese Journal of Pop-
ulation, Resources and Environment 20 (3), pp. 295–306. doi: 10.1016/J.CJPRE.2022.
09.011.
Asadullah, M. Niaz and Saizi Xiao (2020). “The changing pattern of wage returns to edu-
cation in post-reform China”. In: Structural Change and Economic Dynamics 53, pp. 137–
148.doi:10.1016/J.STRUECO.2020.01.010.
98
Chen, Jie and Francesco Pastore (2021). “”Study Hard and Make Progress Every Day”:
Updates on Returns to Education in China”. In: IZA Discussion Paper Series No. 14139.
Gao, Xing and Maishou Li (2022). “Differences between returns to education in Urban
and rural China and its evolution from 1989 to 2019”. In: PLOS ONE 17 (10), e0274506.
doi:10.1371/JOURNAL.PONE.0274506.
Atinc, Tamar Manuelyan (1997). “Sharing Rising Incomes: Disparities in China”. In.
Yang, Dennis Tao (1999). “Urban-Biased Policies and Rising Income Inequality in China”.
In: AEA Papers and Proceedings 89 (2), pp. 306–310.
Akg¨ uc ¸, Mehtap, Corrado Giulietti, and Klaus F Zimmermann (2013). “The RUMiC Lon-
gitudinal Survey: Fostering Research on Labor Markets in China”. In: IZA Discussion
Paper No. 7860.
Kong, Sherry Tao (2010). “Rural–Urban Migration in China: Survey Design and Imple-
mentation”. In: The Great Migration. Ed. by Xin Meng, Chris Manning, Li Shi, and Tad-
juddin Nur Effendi. Chapters. Edward Elgar Publishing. Chap. 7.
Mincer, Jacob (1974). Schooling, Experience and Earnings. ISBN: 0-87014-265-8. National Bu-
reau of Economic Research, New York.
Blattman, Christopher and Stefan Dercon (2018). “The impacts of industrial and entrepreneurial
work on income and health: Experimental evidence from Ethiopia”. In: American Eco-
nomic Journal: Applied Economics 10.3, pp. 1–38.
Banerjee, Abhijit V and Esther Duflo (2014). “Do firms want to borrow more? Testing
credit constraints using a directed lending program”. In: Review of Economic Studies
81.2, pp. 572–607.
99
De Mel, Suresh, David McKenzie, and Christopher Woodruff (2019). “Labor drops: Exper-
imental evidence on the return to additional labor in microenterprises”. In: American
Economic Journal: Applied Economics 11.1, pp. 202–235.
Blattman, Christopher, Eric P Green, Julian Jamison, M Christian Lehmann, and Jeannie
Annan (2016). “The returns to microenterprise support among the ultrapoor: A field
experiment in postwar Uganda”. In: American economic journal: Applied economics 8.2,
pp. 35–64.
Mckenzie, David J and Susana Puerto (2017). “Growing markets through business train-
ing for female entrepreneurs: A market-level randomized experiment in Kenya”. In:
World Bank Policy Research Working Paper 7993.
Atkin, David, Azam Chaudhry, Shamyla Chaudry, Amit K Khandelwal, and Eric Ver-
hoogen (2017). “Organizational barriers to technology adoption: Evidence from soccer-
ball producers in Pakistan”. In: The Quarterly Journal of Economics 132.3, pp. 1101–1164.
McKenzie, David (2017). “Identifying and spurring high-growth entrepreneurship: Ex-
perimental evidence from a business plan competition”. In: American Economic Review
107.8, pp. 2278–2307.
Hardy, Morgan and Gisella Kagy (2020). “It’s getting crowded in here: experimental evi-
dence of demand constraints in the gender profit gap”. In: The Economic Journal 130.631,
pp. 2272–2290.
— (2018). “Mind the (profit) gap: why are female enterprise owners earning less than
men?” In: AEA Papers and Proceedings. Vol. 108, pp. 252–55.
Hjort, Jonas and Lin Tian (2021). The economic impact of internet connectivity in developing
countries. INSEAD.
100
Dolfen, Paul, Liran Einav, Peter J Klenow, Benjamin Klopack, Jonathan D Levin, Larry
Levin, et al. (2023). “Assessing the gains from e-commerce”. In: American Economic
Journal: Macroeconomics 15.1, pp. 342–370.
Luo, Xubei and Chiyu Niu (2019). “E-Commerce participation and household income
growth in Taobao Villages”. In: World Bank Policy Research Working Paper 8811.
Couture, Victor, Benjamin Faber, Yizhen Gu, and Lizhi Liu (2021). “Connecting the coun-
tryside via e-commerce: evidence from China”. In: American Economic Review: Insights
3.1, pp. 35–50.
Bhuller, Manudeep, Domenico Ferraro, Andreas R Kostøl, and Trond C Vigtel (2023). The
internet, search frictions and aggregate unemployment. Tech. rep. National Bureau of Eco-
nomic Research.
Geraci, Andrea, Mattia Nardotto, Tommaso Reggiani, and Fabio Sabatini (2022). “Broad-
band internet and social capital”. In: Journal of Public Economics 206, p. 104578.
Balgobin, Yann and Antoine Dubus (2022). “Mobile phones, mobile Internet, and employ-
ment in Uganda”. In: Telecommunications Policy 46.5, p. 102348.
Zuo, George W (2021). “Wired and hired: Employment effects of subsidized broadband
Internet for low-income Americans”. In: American Economic Journal: Economic Policy
13.3, pp. 447–482.
Hjort, Jonas and Jonas Poulsen (2019). “The arrival of fast internet and employment in
Africa”. In: American Economic Review 109.3, pp. 1032–79.
Yang, Mengjun, Shilin Zheng, and Lin Zhou (2022). “Broadband internet and enterprise
innovation”. In: China Economic Review 74, p. 101802.
101
Lin, Chen, Chicheng Ma, Yuchen Sun, and Yuchen Xu (2021). “The telegraph and modern
banking development, 1881–1936”. In: Journal of Financial Economics 141.2, pp. 730–749.
Xu, Xu, Alison Watts, and Markum Reed (2019). “Does access to internet promote inno-
vation? A look at the US broadband industry”. In: Growth and Change 50.4, pp. 1423–
1440.
Xie, Mengmeng, Lin Ding, Yan Xia, Jianfeng Guo, Jiaofeng Pan, and Huijuan Wang (2021).
“Does artificial intelligence affect the pattern of skill demand? Evidence from Chinese
manufacturing firms”. In: Economic Modelling 96, pp. 295–309.
Chen, Shiyi, Wanlin Liu, and Hong Song (2020). “Broadband internet, firm performance,
and worker welfare: Evidence and mechanism”. In: Economic Inquiry 58.3, pp. 1146–
1166.
Cusolito, Ana Paula, Daniel Lederman, and Jorge Pe˜ na (2020). “The effects of digital-
technology adoption on productivity and factor demand”. In.
Chen, Lipeng and Wanlin Liu (2022). “The effect of Internet access on body weight: Evi-
dence from China”. In: Journal of Health Economics 85, p. 102670.
Niu, Geng, XiaoShu Jin, Qi Wang, and Yang Zhou (2022). “Broadband infrastructure and
digital financial inclusion in rural China”. In: China Economic Review 76, p. 101853.
Leng, Xuan (2022). “Digital revolution and rural family income: Evidence from China”.
In: Journal of Rural Studies 94, pp. 336–343.
Viollaz, Mariana and Hernan Winkler (2022). “Does the internet reduce gender gaps? The
case of Jordan”. In: The Journal of Development Studies 58.3, pp. 436–453.
102
Appendices
F Empirical Strategy in Lemieux 2006b
The paper (Lemieux 2006b) extends standard human capital pricing model in the litera-
ture to a random coefficient model incorporating heterogeneous return to education and
experience. The labels of equations are consistent with (Lemieux 2006b).
w
it
=α
t
a
i
+f (S
i
,β
t
)b
i
+g (X
i
,γ
t
)c
i
+e
it
(2)
wheref (·) andg (·) are of different specification in different versions of the model, e
it
∼
i.i.d.
0,σ
2
t
is an idiosyncratic error. In the baseline case, the specification involves lin-
ear spine in education,f (S
i
,β
t
) =β
1t
S
i
+β
2t
S
i
1(S
i
> 12), and quadratic specification for
experience,g (X
i
,γ
t
) =
P
4
q=1
γ
qt
X
q
i
. The key implication of the model (2) is delivered by
conditional mean and variance of wages under the assumption that random coefficients
a
i
,b
i
andc
i
are uncorrelated and have unit mean by normalization:
E(w
it
|S
i
,X
i
) =α
t
+f (S
i
,β
t
) +g (X
i
,γ
t
) (3)
Var(w
it
|S
i
,X
i
) =α
2
t
σ
2
a
+f (S
i
,β
t
)
2
σ
2
b
+g (X
i
,γ
t
)
2
σ
2
c
+σ
2
t
(4)
The conditional moments of wages clearly show that an increase in price of different
levels of education will not only increase the level of wages but also the dispersion of
wages. However, we cannot estimate Eq (3) and (4) directly. The strategy of the paper
103
is to divide wages into predicted and residual components,w
it
=p
it
+r
it
, by a regression
of log wages on a full set of education and experience dummies and interactions between
nine schooling dummies and a quartic in experience, as in (Lemieux 2006a).
w
it
=
X
s
˜
β
st
S
ist
| {z }
education dummies
+
X
k
˜ γ
kt
X
ikt
| {z }
experience dummies
+
X
s
ξ
st
S
ist
X
4
i
| {z }
interactions
+δ
it
(P)
p
it
is the predicted wage obtained by (P) andr
it
is the regression residual. Then we
can jointly estimate
p
it
=a
t
+f (S
i
,β
t
) +g (X
i
,γ
t
) +u
it
(5)
r
2
it
=α
2
t
σ
2
a
+f (S
i
,β
t
)
2
σ
2
b
+g (X
i
,γ
t
)
2
σ
2
c
+σ
2
t
+v
it
(6)
Note thatp
it
andr
2
it
are approximations of E(w
it
|S
i
,X
i
) and Var(w
it
|S
i
,X
i
) in (3) and
(4), respectively. The error termsu
it
andv
it
in Eq (5) and (6) are approximation errors.
The sources of the error are estimation error generated when estimating (P) and model
difference between (2) and (P).
In particular, if we focus onu
it
, the source of theu
it
is the difference between the right-
hand side of (P) and (3). If the right-hand side of (P) and (3) are the same,u
it
should be
identical to 0. If the argument up to now makes sense, we can decomposeu
it
andv
it
into
two parts,
u
it
=ε
(u)
it
+ ˜ u
it
, v
it
=ε
(v)
it
+ ˜ v
it
whereε
(u)
it
andε
(v)
it
are idiosyncratic errors and ˜ u
it
and ˜ v
it
are due to model differences.
Though we don’t have the explicit form ˜ u
it
and ˜ v
it
, it is intuitive that they are dependent
onX
i
andS
i
. Recall that in the simplest model misspecification example omitted variable
bias, the bias is dependent on regressors.
104
Consequently, (5) and (6) encounters endogeneity problem sinceu
it
andv
it
are cor-
related withf andg. This problem will appear regardless of what kinds of structure we
impose on (P), as long as it is different from (3) or (5). (Lemieux 2006b) tends to ignore
the endogeneity problem and direct estimates Eq (5) and (6).
105
Abstract (if available)
Linked assets
University of Southern California Dissertations and Theses
Conceptually similar
PDF
Essays on macroeconomics and income distribution
PDF
Essays on the dual urban-rural system and economic development in China
PDF
Three essays on the microeconometric analysis of the labor market
PDF
Essays on applied microeconomics
PDF
Essays on labor and development economics
PDF
Essays on development and health economics: social media and education policy
PDF
Entrepreneurial labor: digital work and subjectivities in China’s new economy
PDF
Essays on development economics and adolescent behavior
PDF
Essays on treatment effect and policy learning
PDF
Growth and development in Africa: macroeconomic and microeconomic perspectives
PDF
Three essays on health economics
PDF
Essays on education and institutions in developing countries
PDF
The impact of economic shocks on firm behavior: insights from three studies
PDF
KAT5 acetyltransferase in DNA damage repair in Schizosaccharomyces pombe
PDF
Essays on development and health economics
PDF
The EITC, labor supply, and child development
PDF
The impact of minimum wage on labor market dynamics in Germany
PDF
Essays in macroeconomics and macro-finance
PDF
Essays on policies to create jobs, improve health, and reduce corruption
PDF
Intergenerational transfers & human capital investments in children in the era of aging
Asset Metadata
Creator
Li, Xiongfei
(author)
Core Title
Essays in development economics
School
College of Letters, Arts and Sciences
Degree
Doctor of Philosophy
Degree Program
Economics
Degree Conferral Date
2023-05
Publication Date
05/11/2023
Defense Date
04/12/2023
Publisher
University of Southern California
(original),
University of Southern California. Libraries
(digital)
Tag
Chinese labor market,education returns,gender gaps,income inequalities,OAI-PMH Harvest,self-employment
Format
theses
(aat)
Language
English
Contributor
Electronically uploaded by the author
(provenance)
Advisor
Strauss, John (
committee chair
), Bairoliya, Neha (
committee member
), Bennett, Daniel (
committee member
), Weaver, Jeffrey (
committee member
)
Creator Email
johnli0328@gmail.com,xiongfel@usc.edu
Permanent Link (DOI)
https://doi.org/10.25549/usctheses-oUC113120738
Unique identifier
UC113120738
Identifier
etd-LiXiongfei-11825.pdf (filename)
Legacy Identifier
etd-LiXiongfei-11825
Document Type
Dissertation
Format
theses (aat)
Rights
Li, Xiongfei
Internet Media Type
application/pdf
Type
texts
Source
20230511-usctheses-batch-1042
(batch),
University of Southern California
(contributing entity),
University of Southern California Dissertations and Theses
(collection)
Access Conditions
The author retains rights to his/her dissertation, thesis or other graduate work according to U.S. copyright law. Electronic access is being provided by the USC Libraries in agreement with the author, as the original true and official version of the work, but does not grant the reader permission to use the work if the desired use is covered by copyright. It is the author, as rights holder, who must provide use permission if such use is covered by copyright.
Repository Name
University of Southern California Digital Library
Repository Location
USC Digital Library, University of Southern California, University Park Campus MC 2810, 3434 South Grand Avenue, 2nd Floor, Los Angeles, California 90089-2810, USA
Repository Email
cisadmin@lib.usc.edu
Tags
Chinese labor market
education returns
gender gaps
income inequalities
self-employment