Tilburg University
Essays on labor and family economics in China
Lei, Lei DOI: 10.26116/center-lis-2002 Publication date: 2020 Document Version
Publisher's PDF, also known as Version of record
Link to publication in Tilburg University Research Portal
Citation for published version (APA):
Lei, L. (2020). Essays on labor and family economics in China. CentER, Center for Economic Research. https://doi.org/10.26116/center-lis-2002
General rights
Copyright and moral rights for the publications made accessible in the public portal are retained by the authors and/or other copyright owners and it is a condition of accessing publications that users recognise and abide by the legal requirements associated with these rights. • Users may download and print one copy of any publication from the public portal for the purpose of private study or research. • You may not further distribute the material or use it for any profit-making activity or commercial gain
• You may freely distribute the URL identifying the publication in the public portal
Take down policy
If you believe that this document breaches copyright please contact us providing details, and we will remove access to the work immediately and investigate your claim.
in China
Proefschrift
ter verkrijging van de graad van doctor aan Tilburg University op gezag van de
rector magnificus, prof. dr. K. Sijtsma,
in het openbaar te verdedigen ten overstaan van een
door het college voor promoties aangewezen commissie
aan Tilburg University op dinsdag 26 mei 2020 om 16.00 uur door
Lei Lei
Prof.dr. J.H. Abbring
Promotiecommissie: Prof.dr. C. Nicoletti
First and foremost, I would like to express my most sincere gratitude to my supervisors, Prof. Arthur van Soest and Prof. Jaap Abbring. They have provided constant support and academic guidance in all the stages of my Ph.D. period. Arthur has always encouraged me to take out my own ideas and help polished them into completed projects by commenting on various versions of proposals, asking insightful questions, and indicating possible solutions when I get lost. Jaap advises me with his sharp intuition in economics and exceptional sense of humor. Being their student provides me precious opportunity to learn how to become an independent researcher and extraordinary mentor. I am deeply indebted to Bettina. She has provided tremendous support when I worked on my job market paper, and invaluable advice when I went through the job market. I also would like to extend my sincere gratitude to the other committee members, Cheti, Pilar, and Martin. Their constructive comments and suggestions have substantially improved this dissertation, and will continue have substantial impacts on my following work.
I am thankful to many excellent colleagues in the department of Econometrics and Operations Research. I am grateful to Bart, Jaap and Tobias for organizing the weekly Structural Econometrics Group lunching meeting. It is a wonderful platform where we can share work in progress and get stimulating feedback from both senior researchers and fellow Ph.D. students.
Great thanks go to our secretaries, Anja Manders-Struijs, Anja Heijeriks, Heidi van Veen, Monique Mauer, management assistant Korine, and graduate officer Cecile for their impeccable assistance and organizations, especially during my job search process.
Besides, my thanks go to my dears friends and colleagues: I am thankful to my classmates Yadi, Shuai, Nickolas, Elisabeth, Emanuel, Yuyan, Bohan, Bowen, Qiuli, Yeqiu, Clemens, Abhilash, and Manuel. I want to thank Emanuel, Renata, Johan, Kadircan, and Yi, who are wonderful office mates. I want to thank Tao, Linbo, Xiaoyu, and Carlos for all the topics we had talked genuinely and food we explored together. I am also deeply grateful to: Hanan Ahmed, Ruonan Fu, Rafael Greminger, Chen He, Yi He, Maciej Husiatyński, Vijaya Kedari, Xu Lang, Manwei Liu, Zihao Liu, Shuang Lv, Xiaoyin Ma, Laura Capera Romano, Ittai Shacham, Suraj Upadhyay, Oliver Wichert, Mingjia Xie, Yan Xu, Yifan Yu, Wanqing Zhang, Nan Zhao, Jianzhe Zhen, Kun Zheng, and Bo Zhou.
I would like to express my deepest gratitude to Zilong. Thank you for always having faith in me, cheering me up when I am down, and sharing joys and sorrows in life.
1 Introduction 1
2 Childcare Choice and the Long-Run Human Capital Outcomes in China 3
2.1 Introduction . . . 3
2.2 Literature on Childcare and Children Outcomes & China Background . . . 7
2.2.1 Early Childhood Education in China . . . 9
2.2.2 Childcare by Grandparents in China . . . 11
2.3 Data Description . . . 12
2.3.1 CHARLS data . . . 12
2.3.2 Variables of Formal and Informal Childcare . . . 13
2.3.3 Measure of Children’s Labor Market Outcomes in Adulthood . . . 14
2.3.4 Control Variables . . . 15
2.4 Empirical Methods and Results . . . 16
2.4.1 Identification Method: IV/2SLS . . . 16
2.4.2 Threat to Identification . . . 18
2.4.3 Baseline Estimation Results . . . 21
2.5 Robustness Check . . . 23
2.5.1 Additional Community History Information . . . 23
2.5.2 Validity of the Instrument Survival Status of Paternal Grandmother . . . . 24
2.5.3 Alternative Measure of Grand-parental Care . . . 25
2.5.4 Heterogeneous Effects . . . 26
2.6 Potential Mechanisms . . . 29
2.6.1 The Role of Grandparents’ Son Preference . . . 29
2.6.2 Change in Quantity of Children . . . 30
2.6.3 Maternal Labor Supply . . . 31
.1 Additional Tables for Chapter 2. . . 48
3 The Effect of Prospective Fertility on Female Labor Market Outcomes: Evidence from a Birth Quota Change in China 71 3.1 Introduction . . . 71
3.2 Related Literature . . . 75
3.2.1 Economic Theories of Discrimination . . . 75
3.2.2 Empirical Approach of Testing for Discrimination . . . 76
3.2.3 Motherhood Penalty . . . 77
3.2.4 The Family Plan Policy and Female Labor Market Conditions . . . 78
3.3 Identification Design . . . 80
3.4 Data and Main Variables . . . 82
3.5 Empirical Results and Discussion . . . 85
3.6 Conclusion . . . 89
3.7 Tables . . . 89
.1 Additional Tables and Survey Questions for Chapter 3 . . . 99
4 The Spillover Effect of Sibling Education on Own Education and Health in China 107 4.1 Introduction . . . 107
4.2 Literature Review . . . 110
4.3 Compulsory Education Law . . . 111
4.4 Data and Variables . . . 114
4.4.1 Data and Sample . . . 114
4.4.2 Siblings’ Education and Birth Information . . . 115
4.4.3 Measures of Health . . . 116
4.4.4 Measures of Health Behavior. . . 117
4.5 Research Design . . . 118
4.6 Main Results . . . 119
4.6.1 First Stage. . . 119
4.6.2 Education Spillover Effect among Siblings . . . 121
4.6.3 Effect of Sibling Education on Health & Health Behavior . . . 122
2-1 Birth Year Distribution of Children Cohorts . . . 35
2-2 Probability of having kindergartens in the community over time for Urban and Rural areas . . . 35
2-3 Distribution of Informal childcare for Urban and Rural Registration Status . . . 36
2-4 Kindergarten Enrolment Rate for Urban and Rural Registration Status . . . 36
2-5 Living Place Choice When Married: By Gender and Time . . . 37
2-6 Event Coefficients Around the Birth of the First Child : By Child Gender . . . 45
2-7 The Childcare Effects on the Maternal Labor Supply: Girls . . . 46
2-8 The Childcare Effects on the Maternal Labor Supply: Boys . . . 47
4-1 Birth Years of Individuals with No Siblings . . . 126
4-2 Exposure to CEL in 1985: by Birth Year . . . 126
2.1 Descriptive statistics of variables . . . 38
2.2 Community-FE IV regression: First Stage Results . . . 40
2.3 Community-FE IV Regression for Adult Labor Market Outcomes: By Child Gender 41 2.4 Community-FE OLS Regression for Adult Labor Market Outcomes: by Child Gender 42 2.5 Community-FE IV Regression: Son Preference of Grandmothers . . . 43
2.6 Community-FE IV Regression for Chance of Having Younger Siblings . . . 44
A1 Reduced Form Regression for Education: Comparing Kindergarten Availability Under Age 6 and 6 Years Later . . . 48
A2 Reduced Form Regression for Father’s Chance of Work . . . 49
A3 Robustness Check for Girls: Years of Education . . . 50
A4 Robustness Check for Girls: Job Prestige . . . 51
A5 Robustness Check for Girls: Job Professional Level . . . 52
A6 Robustness Check for Boys: Years of Education . . . 53
A7 Robustness Check for Boys: Job Professional Level . . . 54
A8 Check the Validity of Instrument Survival Status of Paternal Grandmother . . . 55
A9 Drop Children Whose Grandmothers died between age 0 and 6 . . . 56
A10 Robustness Check by Using The Provision of Grand-parental Care between Age 3 and 5: The First-Stage Results . . . 57
A11 Robustness Check by Using The Provision of Grand-parental Care between Age 3 and 5: The Second-Stage Results . . . 58
A12 Heterogeneous Effect by Literacy of grandmother: Girls . . . 59
A13 Heterogeneous Effect by Literacy of grandmother: Boys . . . 60
A14 Heterogeneous Effect by Education of Mother: Girls . . . 61
A15 Heterogeneous Effect by Education of Mother: Boys . . . 62
A16 Heterogeneous Effect by Urban Registration: Girls . . . 63
A19 Heterogeneous Effect by Sibling Sex Composition: Boys . . . 66
A20 Heterogeneous Effect by Birth Order: Girls . . . 67
A21 Heterogeneous Effect by Birth Order: Boys . . . 68
A22 Robustness Check for Girls: Chance of Maternal Subsequent Childbearing . . . 69
A23 Robustness Check for Boys: Chance of Maternal Subsequent Childbearing. . . 70
3.1 The control and treatment groups . . . 81
3.2 The Birth Quota Change . . . 81
3.3 The definition of variables . . . 90
3.4 Descriptive Statistics of Variables . . . 91
3.5 Fertility Effect of 2013 Policy . . . 93
3.6 DID Regression for Married Women: Chance of Having a Job . . . 93
3.7 DID Regression for Married Women: Chance of Having a Non-Agricultural Job . . 94
3.8 DID Regression for Married Women: Monthly Working Hours . . . 94
3.9 DID Regression for Married Women: Hourly Wage . . . 95
3.10 DID Regression for Married Women: Monthly Wage . . . 95
3.11 DID regression for Married Women: Heterogeneous effect Over the Age When 1st Child Was Born . . . 96
3.12 DID regression for Married Women: Heterogeneous effect Over Fertility Desire . . 96
3.13 DID regression for Married Women: Heterogeneous effect Over Years on Current Job. . . 97
3.14 DID regression for Married Women: Agricultural and Self-Employed Workers. . . 98
3.15 DID regression for Married Men: Chance of Having a Job . . . 99
3.16 DID regression for Married Men: Chance of Having a Non-Agricultural Employed Job. . . 99
3.17 DID regression for Married Men: Monthly Working Hours . . . 100
3.18 DID regression for Married Men: Hourly Wage . . . 100
B1 DID regression for Older Women: Chance of Having a Job . . . 100
B2 DID regression for Older Women: Chance of Having a Non-Agricultural Employed Job. . . 101
B3 DID regression for Older Women: Monthly Working Hours . . . 101
B6 Robustness Check at the Intensive Margin for Married Women: Hourly Wage . . . 102
B7 Robustness Check at the Intensive Margin for Married Women: Monthly Wage . . 102
B8 The definition of job category variable jobtype in CFPS2012 . . . 104
4.1 CEL Implementation Year . . . 128
4.2 Years of Exposure to CEL: Younger Siblings. . . 129
4.3 Years of Exposure to CEL: Older Siblings . . . 129
4.4 Descriptive statistics of variables . . . 130
4.5 First Stage Estimates: CEL Effect on Years of Education . . . 133
4.6 Spillover Effect of Younger Sibling Education on Education and Health . . . 134
4.7 Spillover Effect of Older Sibling Education on Education and Health . . . 135
4.8 Spillover Effect of Younger Sibling Education on Health Behavior . . . 136
4.9 Spillover Effect of Older Sibling Education on Health Behavior . . . 137
4.10 Spillover Effect of Younger Sibling Education on Education and Health: Cohorts 1965-1990 . . . 138
4.11 Spillover Effect of Younger Sibling Education on Health Behavior: Cohorts 1965-1990 . . . 139
4.12 Spillover Effect of Older Sibling Education on Education and Health: Cohorts 1965-1990 . . . 140
4.13 Spillover Effect of Older Sibling Education on Health Behavior: Cohorts 1965-1990141 4.14 Spillover Effect of Younger Sibling Education on Health: by Sibling’s Gender . . . 142
4.15 Spillover Effect of Younger Sibling Education on Health Behavior: by Sibling’s Gender. . . 142
4.16 Spillover Effect of Younger Sibling Education on Health: by Own Gender . . . 142
4.17 Spillover Effect of Younger Sibling Education on Health Behavior: by Own Gender 143 4.18 Spillover Effect of Younger Sibling Education on Health: by Same Gender or Not . 143 4.19 Spillover Effect of Younger Sibling Education on Health Behavior: by Same Gender or Not . . . 143
4.23 Spillover Effect of Younger Sibling Education on Health Behavior: by the Number
Introduction
This thesis contains three chapters that investigate topics lying in the intersection between household and labor economics, with a special focus on the context of China. Chapter 2 studies the long-term effect of childcare choice on children’s education and labor market outcomes. Chapter 3 examines the immediate effect of the family planning policy relaxation in 2013 on female labor supply and explores potential explanations behind the observed effect. Chapter 4 studies how a sibling’s education affects an individual’s own educational and health outcomes, by exploiting the compulsory education law reform around 1986.
Chapter 2 estimates the long-run effects of informal childcare, provided by grandparents, and formal childcare, provided by kindergarten, on human capital outcomes in China. To correct for the endogeneity of childcare choices, I adopt an instrumental variables approach: I exploit the exogenous variation in the survival status of the paternal grandmother when the child was under age 6 to create instruments for the use of grand-parental care, and the existence of community kindergarten in the neighborhood as an instrument for formal care. Using retrospective data provided by parents on childcare choice in early childhood, I robustly find that both types of childcare lead to higher educational attainment and better job outcomes. Moreover, the effect of grand-parental childcare is stronger for girls, consistent with son preference: girls get less care from parents who in general favor boys, and there is a greater scope of improvement for girls if they could receive additional care from grandparents. In other words, grand-parental care helps reduce gender inequality at the starting point of life. To further test the hypothesis, I leverage up the son preference attitude of the grandmother from mother’s side, and use it as a proxy for mother’s son preference. I find the impact of grand-parental care increases with mother’s son preference level. As for the positive long-run impact of formal care, I find that the channel operates through mother’s labor supply and family structure. In particular, it helps mothers work longer hours and
get a higher income, and reduces the family size. Both channels ensure that kids receive sufficient resource and educational investment. My findings highlight the important role of grandparents and public childcare for children in under-developed areas.
Chapter 31exploits a family plan policy change in December 2013, which allowed couples in
which either side is an only child to have two children without any penalty. Using a Difference-in-Difference approach and individual level data collected in 2010, 2012 and 2014, this paper investigates how the policy change affected the labor market position of prime-age married women in three dimensions: participation in paid work, working hours and monthly earnings. This paper finds that the policy increased paid work of treated women by 15 to 17 hours per month, while it had no significant effect on monthly earnings. No effects of the policy change are found on women in the agricultural sector, on self-employed women, or on male workers. Furthermore, having the first child at a later age or a longer tenure on the job reduces the size of the effect on hours worked. The effect of the policy change does not vary with the women’s desired number of children, a variable that is usually unknown to the employer. The evidence as a whole suggests that employers use observable productivity-related characteristics to screen women, since they expect that women who are more likely to have children after the policy change will invest less in their job.
Chapter 4 studies the spillover effect of sibling’s education on one’s own education, health, and health behavior. I use the introduction of compulsory schooling law around 1986 in China as exogenous variation in sibling’s years of schooling, exploiting the fact that the policy effect varies across children born in separate calendar years. I find positive sibling spillover effects in education as well as in health and health behavior. Further heterogeneity analyses provide suggestive evidence in favor of mechanisms through sibling interaction and health information transmission, in addition to through the positive spillover in education. I do not find evidence that parents change investment in children if their siblings received more education by the CEL. This non-negligible externality of education policy suggests using education policy as an instrument to improve population health is more effective than we used to think.
The data and policy changes used in this thesis come from China. It is not only because I grew up there and have observed social phenomenons and have reflected on these for a while. The Chinese settings also provide special perspectives to answer important questions discussed in this thesis, which I will explain explicitly in each chapter.
Childcare Choice and the Long-Run Human
Capital Outcomes in China
1
2.1
Introduction
Early childhood education has attracted tremendous attention as a potential instrument to improve family and child development. On the one hand, available non-parental childcare can increase
parents’, especially mothers’, labor force participation (Gelbach, 2002; Lefebvre & Merrigan,
2008; Berlinski & Galiani, 2007). On the other hand, there is ongoing discussion on whether
non-parental care could be harmful to children or not. There is a growing body of literature that examines how the provision of public childcare might have impacts on both cognitive and non-cognitive skill development for the affected children in developed countries, both in the short
and long run (Baker, Gruber, & Milligan, 2008; Gupta & Simonsen,2010;Felfe & Lalive,2018;
Havnes & Mogstad, 2011). Research on informal care by grandparents usually find the adverse
effect, especially for some measures of no-cognitive skills, in the short run (Del Boca, Piazzalunga,
& Pronzato,2018)2. Little is known whether these adverse effects last into adult life.
Whether involvement of alternative childcare would lead to as good outcomes for children as parental care is an important question to answer, as they provide critical guidance on evaluating
1This chapter is adapted from the paper with the same title. I am grateful to my supervisors Arthur van Soest and Jaap Abbring for their constant support and constructive suggestions. I thank Bettina Siflinger, Cheti Nicoletti, Pilar Garcia Gomez, Martin Salm, Kathleen Mullen, Erik Plug, and seminar participants at Tilburg University. This paper uses data from the China Health and Retirement Longitudinal Study (CHARLS). The CHARLS was supported by the behavioral and Social Research division of the National Institute on Ageing of the National Institute of Health (Grant Nos. 1-R21-AG031372-01, 1-R01-AG037031-01, and 3-R01AG037031-03S1), the Natural Science Foundation of China (Grant Nos. 70773002, 70910107022, and 71130002), the World Bank (Contract Nos. 7145915 and 7159234), and Peking University. All remaining errors are my own.
2In this paper, the baseline mode of care is formal care, so it suggests that comparing to attending childcare center, care by grandparents does worse for children.
policies aiming at improving human capital. Interventions in the early stage of child development
would be more cost-effective than later stage (Heckman,2006;Cunha & Heckman,2007). Studies
about childcare effects in developed countries where parents have higher educational attainment, get married later and in general with fewer children might not necessarily help understand the case in developing countries. One can expect that alternative childcare available, even with mediocre quality, for a child in a society where the average education level of mothers is 4.5 years and age of having the first child is 22, and most people lived in poverty, as the sample I focus on in this paper, does not necessarily lead to a negative effect.
In this paper, I examine the long-term impact of childcare choice in early childhood on childern’s human capital outcome in their later life. I concentrate on a cohort of children born between 1950 and 1995 in China, and study formal and informal childcare separately. The formal care I investigate in this paper is kindergarten attendance, and informal care is the one provided by grandparents. I leverage the data China Health and Retirement Longitudinal Study (CHARLS). In the Life History Data carried out in 2014, interviewed mothers of age 45 or older provide retrospective information on childcare usage when each of their children was under age 6. In wave 2011 and 2013, it ask parents the current job information of their children. I combine these two part data together to answer the question I ask in this paper. As all mothers report that they were one of the main caregivers during a child’s age 0 to 5 in the sample, the estimated effect of alternative childcare is regarded as an additional treatment besides maternal care.
6 years old; 2, it has no correlation with labor market outcome measured by father’s chance of work, which suggests that having a kindergarten does not correlate with local labor market conditions. In
section2.4.3I will discuss the results in more detail.
The human capital outcomes I examine in this paper contain one education measure and five measures on job characteristics. The education measure is the years of schooling. Job characteristics include whether children have a job, children’s household income level, the prestige of children’s job occupation, the professional level and administrative level of their jobs. The household income level of children falls into eleven categories ranging from low to high. The prestige of children’s job occupation ranks occupation from agricultural producers (with assigned value 1) to managers (6), and captures the social prestige of the job children have. Similarly, the professional level (ranging from 0 to 4) and administrative level (ranging from 0 to 5) are also two ordinal variables that describe whether children have a certain level of professional or administrative title on their job.
I find that for girls, on average, attending a kindergarten leads to 0.84 more years of education, and 0.20 higher professional level on the job. The involvement of grandparents increases education
by 3.19 years3 then the prestige level of job by 1.35. For boys, kindergarten leads to 0.83 more
years of education, and 0.15 higher professional level on the job. The involvement of grandparents increases education by 1.87 years, and a professional level by 0.38. These findings reveal an essential gender difference of the childcare effects: childcare by grandparents have a greater educational effect on girls than boys.
My findings are robust to additional community historical characteristics, suggesting the exogeneity of the presence of community kindergarten. I also show that the positive grandparents’ effect is not driven by their education level, and the positive effect is among the children with a disadvantaged family background or an unfavoured position within a family. I argue that the prevalent son preference can explain the greater grand-parenting effect on girls. In particular, if parents invest more resources in boys, grandparents can make up the deficient care girls receive from parents and improve girls’ human capital effectively. In other words, if the production function of human capital has diminishing marginal productivity, then girls benefit because they start from
a worse condition4. Leveraging the son preference level of maternal grandmothers, and using it
3It is worth noting that the IV estimates capture the local average treatment effect (LATE): for grandparents, it captures the impact for children who would receive childcare from grandparents only if their paternal grandmother was still alive back then. As I will see in the first stage regression table in2.2, with an alive paternal grandmother increases the chance of grandchild care by 16.4% (16.6%) for girls (boys), while the presence of paternal grandfather decreases the chance by 9.9% (9.2%) for girls (boys). As a comparison, the presence of a community kindergarten increases the chance of kindergarten attendance by 36.6% (41.4%).
as a proxy variable for maternal gender attitude provides supportive evidence to this hypothesis. For the formal care from kindergarten, I find two mechanisms that can explain the long-run effect: higher maternal labor supply and labor income for girls under age 6 but not for boys; lower maternal chance of having subsequent childbearing and smaller family size. Both mechanisms ensure existing children in the household receiving more investment and accumulate more human capital.
This paper adds to the rapidly growing literature that examines how childcare in early childhood can have short-run and long-run effects. Most research focuses on the form care provision in
developed countries (Berlinski & Galiani, 2007; Baker et al., 2008; Gupta & Simonsen, 2010;
Felfe & Lalive, 2018among others) and most examine the short and middle term effect on child
development (by tracking children into primary and middle school), exceptHavnes and Mogstad
(2011) andHavnes and Mogstad (2015) finding the long-run effect of the expansion of universal
childcare in Norway on education and wages at age 30. Using the instrumental variable approach, I estimate the (local, in the case of heterogeneous treatment effect) average treatment effect (ATE), other than the intent-to-treat effect (ITT) in most previous literature that examines the effectiveness of childcare expansion policies. Although ITT has direct policy relevance in the sense that it is under the control of policymakers, the ATE of kindergarten can provide useful policy guidance for the current ongoing early childcare expansion in China since 2010. I focus on a cohort mainly born in an agrarian economy whose main care provider (mainly mothers) were occupied in agricultural work. With the enhanced female education level and more opportunities in the formal sectors, the demand for other types of childcare is increasing as the employed mothers have less flexibility to balance work and childcare. With a higher expected kindergarten enrolment rate than before, the ATE in my paper provides a better reference to the concerned ITT of current on-going early childcare expansion.
My paper is among the few studies (Del Boca et al., 2018) that examine the child outcomes
from grandmother-provided childcare. In a developing economy where the public investment in the formal childcare is far from satisfaction, the role of available informal care with a mediocre quality can also play a role in child development, as I find here. Also, the finding that grand-parenting improves girls’ education attainment and labor market performance suggests that the policy makers should take this contribution into account when considering the retirement and pension design for elderly people. China has a 5-year earlier statutory retirement age for working women than men,
and it already causes a significant disadvantage in the pension for women (R. Zhao & Zhao,2018).
Besides, this paper is among the few studies which focus on the gender difference in childcare
choice and child development. Literature in psychological development (Magnuson & Duncan,
2016) has noticed that there is a gender difference in the development timing of both cognitive
and non-cognitive skills. Research in developed countries suggests that boys are more sensitive
than girls to childcare intervention in early childhood (Chetty et al., 2011; Havnes & Mogstad,
2011). Garcia, Heckman, and Ziff(2018) find that in the long run girls gain more in the childcare
program ABC/CARE, a random control trial (RCT) of quality early childhood education in the U.S.. The reason behind this gender difference is that girls in the program usually suffer worse home
environment5than boys. It leads to a greater scope for improvement for girls in the program. This
paper considers a society where the son preference is dominating. Thus girls and boys may have an unequal developmental level due to unequal parental resource distribution. I observe that girls gain more considerable educational improvement from grand-parental care. I argue that gender difference in grand-parenting is due to the long-existing son preference in China, where girls get less investment from parents and make the care from grandmothers become more valuable. This
explanation is similar to the one in Garcia et al. (2018). But here girls have less parental care
because parents prefer sons.
This paper is organized as follows. In the following section 2.2, I review the literature in
childcare and long term outcomes and briefly introduce the childcare evolution in China. In section
2.3, I describe the CHARLS data and key variables I use. Section 2.4 I introduce the empirical
models and present the results, followed by the robustness check and heterogeneous effect in section
2.5.Section2.6tests potential mechanisms. In section2.7I discuss the implication of my findings
and conclude the paper.
2.2
Literature on Childcare and Children Outcomes & China
Background
Unlike high-income nations, formal childcare in China is far from universal, and the support from government has experienced a considerable change since the founding of the People’s Republic of China in 1949. The availability of alternative childcare types, on the one hand, can reduce the burden of mothers, who are usually the primary care providers in the household, thus help mothers have a better chance in the labor market. The enhanced maternal labor income can, in turn, improve child development. On the other hand, current research has inconclusive evidence on
whether formal care and informal childcare (Blau & Currie,2006) improve the children’s outcomes in health and education, depending on the substitution in parenting time and care quality between parental care and other types of care.
For the formal childcare, the findings are consistent that it improves maternal labor supply
(Lefebvre & Merrigan,2008in Canada;Berlinski & Galiani,2007in Argentina;Du & Dong,2013
in urban China). Its improvement on children’s development is under debate: Baker et al.(2008)
andGupta and Simonsen (2010) find that universal childcare has no or negative effect on
short-term child development, in Quebec and Denmark, respectively. However, German data analyzed
by Felfe and Lalive(2018) show that early childcare center improves children’s socio-emotional
development. In studies where positive effects of public childcare provision are found, the effect
magnitude is always larger for disadvantaged families (Havnes & Mogstad, 2011). In studies that
discover the detrimental effect, childcare usually starts very early (for children under age 3), and the
quality is not comparable with parental care at home (Baker, Gruber, & Milligan,2015). In terms
of gender difference, which is an important focus in this paper, some studies in developed countries show that boys are more sensitive to environmental changes and are more likely to be negatively
affected by public childcare programs, compared to girls (Gupta & Simonsen, 2010). Del Boca,
Martino, Meroni, and Piazzalunga(2019) provides a comprehensive review of the gender difference
in children’s early education. Current findings on gender differences of early childhood education are mixed and inconclusive, and most research is conducted in developed countries. It would be interesting to look at the gender difference in a developing society where exists son preference and the unequal expected labor market return of education for males and females.
The literature on informal childcare, especially the childcare provided by grandparents, has three branches of research focus: First, the labor supply and health (in the dimensions of physical
health, mental health and cognitive health) effect on grandparents themselves (Arpino & Bordone,
2014,Brunello & Rocco,2016andDi Gessa, Glaser, & Tinker,2016taking the early life experience
of grandparents into account using Europe SHARE data;Xu,2018using CHARLS data in China;
Rupert & Zanella,2018using the instrument of grandparents with first child being female tending
to enter grandparenthood earlier due to younger marriage age for females in the United States;
Frimmel, Halla, Schmidpeter, & Winter-Ebmer, 2017 exploiting the assumption that the arrival
time of the first grandchild is exogenous for grandparents using Austrian data); Second, the labor
supply effect on the middle generation, mainly the mothers (Posadas & Vidal-Fernández,2013in
the United States;Bratti, Frattini, & Scervini,2018exploiting a pension reform in Italy;Y. Li,2017
andY. Chen & Zhang, 2018using the mandatory retirement age as instrument of grandchild care
grandparenting supply in urban China); Third, the impact on grandchildren development (Del Boca
et al., 2018 in the UK; Zeng & Xie, 2014 in rural China). Among this literature, there exist
both demonstration of association and attempts to identify the causal effect of grandparenting
on grandparents’ and maternal labor supply (Zanella, 2017 provides an excellent review). They
consistently find the negative effect on grandparents (mainly grandmothers) labor supply and positive effect on maternal labor supply. On the other hand, the research exploring causal effect of grandparenting on grandparents’ health and child development is much scarce, and most research focus on the protecting role of grandparents for children with disadvantaged backgrounds, for
example, single-parent or ethical minority families (Coall & Hertwig,2011).
Comparing with US and European countries, grand-parenting is more common in China and eastern Asian countries. It is partly because that the Confucianism emphasizes connection across extended families, and also that in a society without sufficient social pension system, elder parents provide child care as a way to exchange the elderly care from their adult children. A related sociological feature is the prevailing co-residence with elderly parents after marriage, which partly,
but not totally, explains the common practice of childcare by grandparents in China. Zeng and Xie
(2014) finds that in rural China, coresident grandparents’ education has an as significant effect as
the parents’ on children’s school attainment, while the non-coresident or deceased grandparents do not have such a schooling effect. However, they do not examine the heterogeneous effect across the gender of children, which could matter in a society where the son preference is overwhelmingly
existing. Sun, Zhang, and Hu (2016) notices that among low-educated mothers, families with a
firstborn boy are more likely to have coresident grandparents than families with a firstborn girl, and further higher maternal labor market attachment. This paper provides causal evidence on the relationship between childcare by grandparents in early childhood and children’s human capital outcomes in later life. I find that even grandparents with no education can make a difference in terms of better educational attainment and job quality, especially on girls, which complements the
findings inZeng and Xie(2014).
2.2.1
Early Childhood Education in China
public care goods largely reduced the unpaid burden of main care providers, i.e., married women. In Mao’s era, both female labor participation rate and gender pay equality ranked top around
the world (Kidd & Meng, 2001; Cook & Dong, 2011). Early childhood education experienced
an enormous increase until the 1960s, when the Cultural Revolution led to massive closure of schools from 1967 to 1977, and kindergartens were with no exceptions due to lack of trained teachers. Starting from 1978, series of reforms have taken place both in rural and urban sectors aiming at introducing market economy to previously planned economic system. In rural areas, the household responsibility system replaced the collective farm system, and the emergence of Township and Village Enterprises (TVE) since the mid-1980s (which were privatized in 1990s) created an opportunity of off-farm industrial employment for rural residents, while it crowded out the public care funding from local governments. At the same time, in urban areas, the central-plan economy in Mao’s era was gradually replaced by the market economy. With the decentralization and privatization of SOEs between 1997 and 2000, the government and employer-provided childcare
endured a decline since the late 1990s (Cook & Dong,2011;Liu, Dong, & Zheng,2010).
In 1979, the summary of minutes of The National Working Conference for Child Care (Central Committee of the Communist Party of China 1979) suggested that the development of early childcare and education should be managed collaboratively by State and local governments. For rural areas, Concerns about Early Childhood Education were implemented by the Ministry of Education in 1983. This document encourages villages to participate in the process of childcare and education promotion, together with higher-level local and state governments. It also forced governments to make a special amount of public expenditure on childcare in rural areas, while raising funding in local communities was recommended as well (the so-called "walking in two legs" guideline)
(L. Zhao & Hu,2008). Plus, the pre-school program usually lasted for one year in rural areas instead
of three years as in urban areas, and it sometimes was administrated by the village primary schools. During this period, early childhood education in China again witnessed an expansion until the late 1990s, when series of reforms in state-owned enterprises took place. The privatization of these previously public sectors led to a sharp decline in the provision of non-monetary compensation, including the closure of the affiliated childcare centers. In general, Kindergarten education is provided with higher quality in urban than in rural areas, besides the duration difference mentioned above. Kindergarten are usually accessible for all children living in the nearby community, except for the employer-provided kindergarten, which is only open for employees. I do not use the employer-provided childcare in the first stage of the instrumental variable approach.
by traditional value on academic achievement in Confucianism and emphasis on collectivism in every level of education in Mao’s era. Until the 2000s, the Ministry of Education highlighted the balanced development of young children in health, science, social, language and arts in the latest
issued document of Guidelines for Kindergarten Education (trial version) (http://www.moe.gov.cn)
(Hu & Szente,2009).
2.2.2
Childcare by Grandparents in China
Concerning informal childcare, as the connection of extended family is close, it is not uncommon that grandparents take care of their young grandchildren, especially for urban young couples, of which both sides are full-time workers. In the CHARLS data, grand-parental care is the primary
informal care: 20.4% of children received care from paternal grandparents6, 5.2% received care
from maternal grandparents, 0.8% got care by relatives, 0.8% got care by siblings, and 0.3% by nannies. All the children are reported to have at least one adult caregivers.
Whether grand-parental care can help children accumulate human capital depends on how much it can complement the parental care. If grand-parental care replaces high-quality parental care, then it might harm child development. In contrast, if parents themselves are on average with low education, and need to spend much time on work, then additional care from grandparents is helpful. Besides, Parents who get children taken care of by grandparents can earn more labor income. The higher income will increase children’s early-stage human capital accumulation, for example, by having more books, obtaining better after-school care, and more opportunities for vacations that help the social and psychological development.
It is worth noting that although in rural China, the co-residence of elderly parents with their adult children, especially sons, are quite common, it does not necessarily imply they provide
grandchild care for sure. As we will see in figure2-3drawn from the mother-reported retrospective
parental (mainly maternal) care. In my data, all children are reported to have mothers who are one of the main caregivers under age 6. Grand-parenting in this paper means that grandparents involved in childcare as primary care providers, in addition to maternal care.
2.3
Data Description
2.3.1
CHARLS data
The data I use in the empirical analysis is CHARLS (China Health and Retirement Longitudinal
Study)7. Is it a representative sample of residents ages 45 and above in China. The baseline
national wave was being fielded in 2011. The baseline sample is drawn by applying 4-stage sampling method. The 4 stages ranging from top to bottom are the count-level sampling, the neighborhood-level sampling, the household-level, and individual-level sampling. The following up waves take place every two years. It contains about 10,000 households and 17,500 individuals in 150 counties/districts and 450 villages/resident communities. In each wave, the children’s information on basic demographic backgrounds, education, and job characteristics are provided by elderly parents, no matter the children live together with the parents or not. Except for the usual waves, a Life History Survey took place in 2014. It is a retrospective survey documenting life events in family, residence, health, wealth, education, and employment of all the CHARLS living respondents in wave 2011 and 2013. The event-history calendar method is adopted to minimize the memory bias and ensure the accuracy of answers. I focus on Children’s Information and Job History in the life history survey. The childcare information (Maternity leave length, kindergarten attendance, and informal care providers) when the child was under age 6 can be found in the Children Information section. I use the job history to identify the employment status, working hours and labor income, and other job characteristics of parents when each child was born (and also between ages 0 and 5). The job history data provides information on the respondent’s first job and any other jobs lasting at least six months. In this paper, as my main dependent variables are the human capital outcomes of the adult children of the CHARLS respondents, I take the perspective of children, referring to the respondents as parents, and the parents of the respondents as grandparents.
In figure 2-1, We can see that the majority of children in my sample were born before 1990. I
restrict to children born between 1950 and 1995 in my final sample.
2.3.2
Variables of Formal and Informal Childcare
In this paper, formal care is attending the kindergarten, and informal care is the childcare provision from grandparents. I include both maternal and paternal grandparents who were reported by parents that they mostly took care of the young child, while in my data, it is primarily the paternal grandparents and secondly the maternal grandparents who helped raise a young child. It is worth noting the way it asks in the CHARLS Life History Survey about the provision of grandchild care: it only asks who were the main caregivers, although it allows multiple choices of involved people,
from parents to paid nannies8. All children had their mothers as one of the primary care provides
in my sample. It implies that reported grand-parenting underestimates help from grandparents to take care of young children, because it excludes grandparents who provide help but do not involve as main caregivers. The share of children whose grandparents involved taking care of them is 26%,
calculated by using parents’ reported retrospective data. Figures2-2and2-3present an overview
of childcare choice in China from 1949 to 2000. As a comparison, I find in CFPS 2014 the share
of grandparents as main caregivers is 30%9. Besides, parents reported that whether there was a
kindergarten present in their living community, or there was a nursery provided by the work unit of parents when the child was between ages 0 and 5. I generate an indicator of whether there is a kindergarten in the community in the following two steps: I first exclude nurseries provided by parents’ workplace. I assume that a kindergarten was present if at least more than 50% of residents reported there was a kindergarten in the same community for children born in the same year. The reason I exclude the nurseries affiliated to parents’ workplace is that before the privatization reform of the state-owned enterprises in late 1990s, a workplace that provided nursery to workers is associated with better work and income for parents. Using this type of formal childcare might pick up the positive effect of parents’ occupation on children’s long-run human capital accumulation. Excluding it helps minimize the correlation of formal care choice and parents’ work characteristics.
From figure 2-2, we can have an overview of how the kindergarten coverage rate at the
community level changed over time for the CHARLS sample. At first, we indeed observe the urban areas have a doubled coverage rate than rural areas. Furthermore, the coverage rate slightly goes up after the reform year 1978. Because the CHARLS data only survey the couples who are 45 or 8The full list of main care givers include parents, paternal grandparents, maternal grandparents, relatives, siblings, nannies and others. As mentioned in section2.2.2, 20.4% of children received care from paternal grandparents, 5.2% received care from maternal grandparents, 0.8% got care by relatives, 0.8% got care by siblings, and 0.3% by nannies. 0.6% were reported as others.
older, I do not have sufficient sample size to estimate the kindergarten rate since the late 1990s, when the decline in government and employer-provided childcare started. It also means that for the children in my sample, they were not affected by this decline in their childhood because they
were already older than six years before the decline started. Figure 2-4 presents the estimated
kindergarten enrolment rate over time in my sample. It keeps low before 1969, and then gradually increases to 15% in rural and 50% in urban areas.
Figure2-3shows that around 70% of families used only parental care. About 16% of families
raise kids under age 3 with grandparents involved. There were more grandparents care and less parental care in urban China, which echoes the fact that urban mothers have more chance of formal and non-flexible employment. Thus, they have a higher demand for help for other care providers (mainly grandparents). The downward shift of parental care in rural areas after 1978 was consistent
with the urbanization and more non-farming industrial work opportunity for rural residents. 10
2.3.3
Measure of Children’s Labor Market Outcomes in Adulthood
I have several measures about human capital outcomes of adult children. Parents as respondents in CHARLS data report the highest education level of each of their child. Household income the child’s household got last year is measured by 11 ordinal interval indicators ranging from 1 representing "No income" to 11 "above 300,000 RMB". High income is defined as the income level is at least 5, where income level 5 indicates annual income between 10,000 and 20,000 RMB, 6 indicates income level between 20,000 and 50,000 RMB. The average level is 5.5. They also report whether the child currently has a job. For children’s primary occupation, I rank it from 1 to 6, where 1 indicates agricultural producers, 2 production and transportation workers, 3 commercial and service workers, 4 clerks, 5 professionals and technicians and 6 managers. I argue that this rank is consistent with occupational prestige in China, which makes sense to interpret the 2SLS results, where I treat occupational prestige as a continuous variable. The professional job level is ranked from 0 with no professional title to 4 with an advanced level. Similarly, the and administrative job level range from 0 with no level to 5 with the director-general of a bureau and above. It is worth noting that around 95% of the children do not have either a professional title or an administrative level.
2.3.4
Control Variables
My control variables contain three parts: the child-specific demographic information, parents’ demographic and economic backgrounds, as well as regional and child cohort fixed effects. The child-level control variables include gender, birth year fixed effect, birth order fixed effect, number
of older brothers11 to capture the parents’ resource allocation based on gender, and whether they
got an urban registration when the child was born. The parents’ information includes maternal age of first birth, both mother and father’s years of education, paternal grandmother’s education level, father’s job classification fixed effect at the initial stage of the job father was occupied when the child was born. I do not add mother’s working information into my control variables. Instead, I
argue change in maternal labor supply as one potential mechanism12: it can be altered by childcare
provision, and , in turn, can have an impact on child human capital involvement (for example,
Baum II, 2003 and Bernal & Keane, 2011). For the regional fixed effect, I use the community
dummies and also a community-specific trend to capture the regional difference that could affect childcare choice and human capital outcomes in adulthood.
Summary Statistics. Table2.1 shows summary statistics for children used in the empirical analysis, for girls and boys respectively. Columns "Not Attend K." and "Attend K." compare children who did not attend a kindergarten and who did so. Similarly, columns "only parents" and "Grandp. involved" compared children who only received parental care and who had grand-parental care. Panel "Parents Characteristics" shows that children attending a kindergarten on average with better home environment: mothers had their first child at later age (23.9 instead of 22), parents are better educated, more likely to have an urban household registration, and with higher labor income. Panel "Child Characteristics" shows that children attending a kindergarten are born later, which is
consistent with the uptrend in figure2-4. They are also more likely to be the one with lower birth
order and has smaller number of siblings. These children are also more likely to live in a community with a kindergarten presence, while the probability of paternal grandparents alive when the child
was under age 6 is not significantly different from children who did not attend kindergartens13, as
seen in the panel "Instrumental Variables". We observe similar pattern for grand-parental care, with one exceptional note: children who got care from grandparents are associated with higher chance that their paternal grandmother is alive in their early childhood (under age 6), but there is 11I control the existing children by gender in the household to address the sex-specific stopping rule: mothers are more likely to stop having more children if they have already had sons. I drop the number of older sisters because it is perfectly linear correlated with the child birth order and number of older brother.
12I test this mechanism in section2.6.3.
not such a association in terms of the survival status of paternal grandfather14. In the following section I describe the empirical method and present the main results.
2.4
Empirical Methods and Results
2.4.1
Identification Method: IV/2SLS
I assume that the long term production function of human capital for children can be written as follows:
yijk= FCijk·γ+ ICijk·η+ Xijk·β+ ChildYearkj+ πk+ τj+ eijk (2.1)
In equation (2.1), yijk is the labor market outcomes of child i born in year j and community k.
FCijk is the formal care choice for child i, and ICijk is the informal care choice for child i in the
family. These childcare choices were made when the child was under age 6. In CHARLS data, more than 99% of non-parental informal care was provided by grandparents. I consider grand-parenting as the only informal care in my study. I take parental care as the baseline choice. In the control variables, I include paternal education attainment and work characteristics for the jobs fathers had when the child was born as a set of proxy for family background. Children’s basic information is also included in the control variable set: an indicator of birth parity, squared age,
and gender. Besides, I add community fixed effect πkand child birth year fixed effect τjto capture
the non-random childcare provision difference and heterogeneity in labor market conditions across
regions and cohorts. I also add the community-specific child birth cohort trendChildYearkjin my
specification, which captures the community-level deviations from the nationwide trend in the birth
year fixed effect (Lundborg & Majlesi,2018). eijkis the error term. I cluster the standard errors at
the community-by-child-birth-year level.
As I pointed out in section2.1, formal care and informal care choice can be correlated with
omitted variables like parental preference and family background that cannot even be absorbed by parents’ education and job characteristics that I control for. This concern leads us to use instruments to account for the endogeneity of formal care and informal care. Moreover, considering that the
gender difference both exists in child sex preference of parents and grandparents (Das Gupta et al.,
2003;Sun et al.,2016) and labor market conditions (L. Zhang, De Brauw, & Rozelle,2004), I split
the sample by the gender of children, and only compare girls with girls and boys with boys. For the formal care, i.e. sending the child to a kindergarten, I use the presence of a kindergarten in the community or village where the child lived up until age 6. Intuitively, the presence of
community kindergarten would increase the chance of attending a kindergarten. It should also hold that the community kindergarten would influence the child later outcomes only through attending the kindergarten. This assumption would be violated if the presence of community kindergarten is correlated with other local economic factors that can impact early childhood development. Then
the legitimacy of this instrument would not hold. To partly address this problem, I add the
community fixed effect to my main specification. That is to say, I apply a difference-in-difference approach, where the control group is the communities that had no kindergarten for all the cohorts of children who lived there. The treated group is the ones that built at least one kindergarten since the born of one cohort of children. The 2SLS estimates the average treatment effect of attending a kindergarten on the treated whose community had a kindergarten in their childhood. With the presence of heterogeneous treatment effect, the 2SLS estimates capture the local average treatment effect (LATE), i.e., it estimates the effect of formal care on human capital outcomes in adulthood only for the children who were sent to the kindergarten only because of the presence of a community kindergarten.
For informal care, i.e., childcare provided by grandparents, I have two instruments. The first one is a binary variable that indicates where the paternal grandmother was still alive during the age 0 to 5 of the child. This instrument should positively correlate with the provision of grand-parenting.
Posadas and Vidal-Fernández(2013) use the death of maternal grandmother as an instrument for
the availability of grand-parental childcare in the United States,Du et al.(2019) use the death of
paternal grandmother instead of the case in China, both with a focus on the effect on maternal labor supply. The idea is that in China, it is paternal grandmothers who help take care of young children, due to the patriarchal tradition that wives after marriage usually moved to husbands’ original house. Also, women are the primary caregivers for the housework. The retrospective data on the marriage history shows that for the parents (the respondents in CHARLS data), 79% of mothers reported that the moved to the husband’s house, while only 5% of fathers said they lived together with the
wife’s house (see figure2-5a). This pattern is consistent over cohorts (see figure 2-5b). What I
do not know exactly is whether the new couple was living in the same household as the husband’s parents. Nevertheless, 20% of children were reported in the sample that their paternal grandparents are main (but not necessarily the only) caregivers under age 3, while only 5% of children were reported that maternal grandparents did so.
The second instrument for grand-parental childcare used in this paper is the interaction between the first instrument and the indicator of the paternal grandfather’s survival status. This instrument should have a negative effect on childcare provision probability by paternal grandmother, in the sense
2018). It means that they also have a burden to take care of their elderly husbands, which creates a conflict for the paternal grandmother between grandchildren’s care and her domestic housework. I also assume that these two instruments influence children later life outcomes only through the childcare that grandparents provided in early childhood. Any violation of this assumption would lead these instruments invalid.
2.4.2
Threat to Identification
I am aware of some concerns that may make the instruments violating independence and exclusion assumptions. One concern is the substitution effect of community kindergarten on other childcare
usages, as found inBaker et al.(2008) in the universal childcare reform in Quebec. If the openness
of community kindergarten leads to less grand-parental care, then the exclusion assumption is violated, and estimating an intent-to-treat effect is preferable. In my case, the use the provision of
childcare by grandparents when the child was under age 315. In the sample, 88% of children who
attended a kindergarten started with age 3 or later, and the average age is 4.2. Besides, I find that in the first stage regression, the presence of a community kindergarten does not change the chance of childcare by grandparents (when the child was under age 3), which provides a piece of evidence that the substitution effect is negligible.
Another concern is whether the existence of community kindergarten is correlated with better access to primary school, or better educational and economic resource in the community. If it is the case, then I should be careful when interpreting the observed positive effect of preschool on the educational attainment of children. I provide four points of argument on the conditional exogeneity of community kindergarten availability. First, access to primary school is relatively high. The national statistics show that in 1958 the enrolment rate of school-age children (no younger than age 7) reached 80.3%, under the government effort in eliminating illiterate and achieving compulsory primary school education at the beginning period of the foundation of People’s Republic of China
since 1949 (Hannum,1999). As a comparison, CHARLS data shown in figure2-4that the estimated
number makes us confident to use the current community as the one for their children under age 6.
In addition, we, in the robustness check section2.5, add a series of community history information,
including: economic regime change, availability of public facilities, whether experience primary school merge, family planning policy and natural disaster events around 6 years before and after the childbirth to indirectly control for the time-varying changes in local economic and governmental development, which might have an impact on the provision of community public kindergarten and child development in the early childhood.My conclusions are quite insensitive to the community level control variables.
Third, I provide evidence from two auxiliary regressions inspired byCurrie and Moretti(2003):
1, I compare the reduced form results on children’s educational attainment, by using kindergarten availability during age 0 to 5 and 6 years later. 2, I check whether the availability of community kindergarten has an impact on the father’s chance of work or not when the child was under age 6. For the first regression, the idea is that if the community kindergarten increases children’s education level by increased their chance of attending a kindergarten, then I would observe that the effect of kindergarten availability during ages 0 to 5 is larger than the one measured 6 years later. For the second regression, if the presence of community kindergarten is a sign of local development, then it is likely to observe that it correlates positively with the labor market performance, measured by the child father’s working status in my case. I use the labor supply of fathers instead of mothers,
because the former is less sensitive to childbirth, as shown in figure 2-6. I discuss the results of
two regressions in section2.4.3.
Fourth, for the potential migration for better public childcare and other educational resources, I argue that it is unlikely to happen in China due to the strict household registration system. This system was introduced in 1958 in Regulations on Household Registration of the People’s Republic of China. The system is used to control population mobility and restrict the distribution of public welfare. People in rural areas only own lands in the place of their household registration and suffer limited public resources in education and health care. Individuals did not have access to food quota, work opportunity, and the public health care system until the end of the 1990s when the government gradually relax the household registration system. Under this system, it is almost impossible for parents to send their young children to a kindergarten that locates in a nearby community.
of paternal grandmothers16, parental education, and father’s job characteristics when the child was born. I experiment with the family fixed effect model, while it ends up with weak instruments: the death of paternal grandmothers under child age 6 does not vary much across siblings in a
household17. I also add the survival status of maternal grandparents as control variable in the
instrumental variable estimation, the results (in section2.5.2) are robust to this change.
The death of grandparents is a permanent shock, and can change children’s outcomes in later life through other channels. For instance, the death of paternal grandmother may change child’s home environment by making the child and other living household members experiencing the feeling of grief. I argue that for children under age 6, providing childcare is the primary channel that a grandparent can directly affect the child development. For older kids who are more likely to develop emotional connection with grandparents, I control for a set of age group fixed effect, of which the age is measured at the year of grandmother’s death. This additional set of variables is aiming at absorbing any direct effect grandmother’s death on child human capital outcomes in later
life other than the indirect effect by providing childcare. Detailed discussion is in section2.5.2.
Besides, when a grandparent stays in bad health condition, he or she may need care from the child’s parents. This may decrease the amount of childcare provided to a child and thus a worse schooling and labor market outcomes in later life. Unfortunately, I do not have the health condition of grandparents during the child’s childhood; thus, I cannot distinguish grandparents who are healthy enough to provide childcare from those who, on the other hand, need elderly care. But given the fact that I exploit the variation of grandmother’s survival status at age 6, 3 years later than the grand-parental care measured in the main result, it is safe to assume that the living grandmother was in a good health condition. I also re-estimate the model by dropping the sample of children whose paternal grandmother passed away when they were between age 0 and 6. The results (in
section2.5.2) are robust to this change.
In addition, being alive of the grandparents can also positively correlate with providing help in housework to the mother, thus encourage mothers to participate in the labor market and increase the household income and the investment in children under age 6. If the channel holds, then I can not interpret my IV estimates as the average treatment effect on the treated in the heterogeneous
effect situation (Angrist & Pischke,2008). However, I argue that taking care of a young child under
age 6 is the most time- and energy-intensive housework, thus childcare is the most likely channel
16I also tried by adding both the education levels of grandmothers and grandfathers, and it does not change my results. These results are not reported in the paper, but available from the author upon request.
that grandparents may affect child development. I discuss these results in more detail in section
2.5.2.
2.4.3
Baseline Estimation Results
First Stage
FCijk = z1ijk·κ+ z2ijk·λ1+ z3ijk·λ2+ Xi·β+ ChildYearkj+ πk+ τj+ µ1ijk (2.2)
ICijk= z1ijk·κ+ z2ijk ·λ1+ z3ijk·λ2+ Xi·β+ ChildYearkj+ πk+ τj+ µ2ijk (2.3)
yijk= z1ijk ·κ+ z2ijk·λ1+ z3ijk·λ2+ Xi·β+ ChildYearkj+ πk+ τj+ µ1ijk (2.4)
In above equations, z1ijk z2ijk and z3ijk represent the 3 instruments: presence of a
commu-nity kindergarten, survival status of paternal grandmother, and the interaction between paternal
grandmother’s survival status and her husband’s. Equation (2.2) is my first stage equation for the
endogenous variable attending a kindergarten when child was under age 6. Equation (2.3) is the first
stage equation for variable grand-parenting. (2.4) is the reduced form equation for my instrumental
variable estimation. I apply Generalized Method of Moments (GMM) to estimate coefficient γ
and η in equation (2.1), which captures the effect of formal and informal care respectively. I first
assume that the error terms in equation (2.2) and (2.3) are independent with the corresponding
independent variables in each equation, and both are correlated with the error term in equation (2.1)
as implied by the endogeneity of childcare choice variables in my main model. GMM estimation also implies that the error terms in the two first stage equations can be correlated with each other, because the moment condition used here is that the correlation between instruments and error term in the second stage equation is zero. I allow cross-section correlation within children born in the same year and in the same community in the stand errors. In other words, I cluster the standard errors at the community-by-child-birth-year level.
In table2.2I present the first stage for the girls’, boys’ and full sample, respectively. My initial
and are less sensitive to the presence of paternal grandfather in terms of getting childcare from grandparents, indicating the prevailing son preference in China. In addition, I notice that the instruments only have an impact on their own endogenous variables, and do not affect the other endogenous variable. Community kindergarten availability only affects the probability of formal childcare provision; Similarly, the two grandparents’ survival status-related instruments only affect the probability of childcare from grandparents. The weak identification test statistics are above
10, the rule-of-thumb threshold for weak instrument test proposed byStaiger, Stock, et al.(1997),
suggested in Baum, Schaffer, and Stillman (2007) for models with robust or clustered standard
error. The under-identification test, weak-identification test, and Anderson-Rubin F test all reject the null hypothesis.
TableA1shows the reduced form regression of children’s years of schooling on instruments and
other exogenous variables for girls’, boys’, and the full sample, respectively, by using kindergarten availability when children under age 6 or between age 6 and 12. For each sample, I find that only the availability under age 6 (in column (2) (4) and (6)) has a significant and positive effect on educational attainment, and using measures 6 years later yield insignificant and much smaller effect. P values in the third row suggest that the effect of having kindergarten under age 6 is significantly different from the effect of having kindergarten between ages 6 and 12, although I can not exclude the possibility that children response with a larger magnitude in an earlier age to the same intervention. Then I turn to the second auxiliary regression.
Table A2 presents the results of the reduced form regression of father’s chance of work on
instruments and other exogenous variables for girls’, boys’, and the full sample, respectively, for the time 1 to 6 years after the childbirth. We can see that overall there is no significant effect of kindergarten presence on father’s work probability, expect for 3 years after the child birth in girls’ sample (the third row in column (1)), and the point estimates are less than 0.01 for most cases. This table suggests that the kindergarten presence is not correlated with local labor market conditions after controlling for community fixed effect, child birth year fixed effect, community-specific trend, and other control variables I use in the main specification.
Human Capital Outcomes in Adulthood
Table2.3 shows the estimation results of human capital outcomes in adulthood for girls and boys
