The effects of formal and informal child care on the Mother's labor supply—Evidence from urban China

China Economic Review 44 (2017) 227–240

Contents lists available at ScienceDirect

China Economic Review journal homepage: www.elsevier.com/locate/chieco

The effects of formal and informal child care on the Mother's labor supply—Evidence from urban China

MARK

Yunrong Li1 School of Public Finance and Taxation, Southwestern University of Finance and Economics, Chengdu, China

AR TI CLE I NF O

AB S T R A CT

Keywords: Informal child care Formal child care Female labor supply

The women's labor force participation rate in China has declined considerably during the last twenty years in urban China. Since the reforms started in the mid-1990s, publicly subsidized child care programs have decreased, and private care centers have increased. This might have increased the reliance of working mothers on informal child care and reduced their reliance on formal child care. Using post-reform data from the Project on Rural–Urban Migration in China (RUMiC) of 2008, I estimate the effects of formal and informal child care on the labor supply of mothers of young children. A recursive model with instrumental variables is employed to account for endogeneity. I find a positive and significant impact of informal child care in the form of grandchild care on the mother's labor force participation, while no significant effect of formal child care in the form of kindergartens or paid nannies. Considering recent tendencies in China to postpone retirement, one possible method to maintain mothers' presence in the labor market could be to reinforce the availability and affordability of formal child care.23

JEL codes J13 J22 H31 1. Introduction Women's labor force participation (LFP) is related with importance issues such as gender equality and household income. Women's LFP rate has decreased in China since the middle of the 1990s. In particular, the LFP rate of women aged between 25 and 49 years old was approximately 91% in 1990, and it decreased to 87.6% in 2000 and to 83.2% in 2010 (Shen, Zhang, & Yan, 2012). Some researchers believe that economic reforms have been a cause of these declines (Yao & Tan, 2005). In view of the stylized inflexibility of balancing motherhood and work, it cannot be ignored that child care uses by the mother, both formal and informal, are important determinants of women's LFP. In urban China, since the launch of a program of radical restructuring of state-owned enterprises in 1997, publicly subsidized

E-mail address: [email protected]. Mailing address: Room 723, Gezhi Building. Liutai Avenue 555, Wenjiang District, Chengdu, Sichuan, P. R. China, 611,130. Abbreviations: RUMiC: Project on Rural-Urban Migration in China. LFP: labor force participation. NBS: China National Bureau of Statistics. cmp: conditional mixed process. 3 Funding: this work was supported by the National Social Science Foundation of China (grant number 15CJY017). The opinions and conclusions expressed herein are solely those of the author and should not be construed as representing the opinions or policies of the foundation or the government. 1 2

http://dx.doi.org/10.1016/j.chieco.2017.04.011 Received 27 October 2016; Received in revised form 30 April 2017; Accepted 30 April 2017 Available online 03 May 2017 1043-951X/ © 2017 Elsevier Inc. All rights reserved.

China Economic Review 44 (2017) 227–240

Y. Li

formal child care programs such as nurseries and kindergartens have significantly decreased, and private care centers offering child care services at higher prices have increased. Since the reforms, public child care programs have been usually available for children aged 3 years old and above, with priority to children whose parents hold local urban registrations (so-called urban Hukou) in current China.4 This might have increased the reliance of working mothers of young children on informal child care and reduced the mothers' reliance on formal child care. Currently, there is a tendency in urban China to delay retiring age. If mothers of young children who receive grandchild care are more likely to participate in the labor force, the delayed retiring age, which might reduce the likelihood of grandchild care, would consequently decrease mothers' LFP. Therefore, the main objective of this paper is to investigate whether grandchild care, referred to as informal child care in this paper, and kindergartens, referred to as formal child care, are important determinants of mothers' labor supply in current China. Previous studies have investigated the relation between different forms of child care and the mother's labor supply. With regard to informal child care, the presence of adult women, especially grandmothers, in the household is generally found to increase the probability of the mother to participate in the labor market (Connelly, Degraff, & Levison, 1996; Del Boca & Vuri, 2005; Du & Dong, 2010, 2013; Kilburn & Datar, 2002; Ogawa & Ermisch, 1996; Sasaki, 2002). Ogawa and Ermisch (1996) and Sasaki (2002) suggest that sharing housework, such as child care, between the young and the elderly avails the young of more time to work at a paid job. However, these studies ignored the non-resident grandparents who may live close enough to provide help. Qu and Sun (2011) note that in urban China the proportion of the elderly who are aged 60 years old or above and live with their children has declined from 56.7% in 2000 to 47.8% in 2006. In addition, the data set used in this paper shows that approximately 23% of preschoolers live only with their parents but are taken care of mostly by their grandparents. If grandchild care helps the mother of young children to participate in the labor force, ignoring nonresident grandparents can lead to an underestimate of the effect of grandchild care on the mother labor supply, as mothers who receive grandchild care from the nonresident elderly will be seen in the analysis as individuals who do not obtain grandchild care. More importantly, if living with grandparents is a result of the mother's decision to work, looking at only resident grandparents will lead to a selection bias that the aforementioned studies have not addressed. There are studies that examine the impact of the presence of both resident and non-resident grandparents on the mother's labor supply. Dimova and Wolff (2008, 2011) studied the case of Europe and found that the provision of grandchild care by grandmothers has a significant positive impact on mother labor supply, but they did not account for formal child care, which might simultaneously affect grandchild care and mother labor supply. Chen, Short, and Entwisle (2000) found that having any grandparent in the household or a paternal grandparent living nearby reduces the mother's involvement in child care in China. Du and Dong (2010) found that having elder women in the household or having grandmothers living nearby has positive impacts on the mother's LFP in China from 1997 to 2006. On the one hand, some of these studies did not consider the possibility of formal child care. On the other, living together or nearby is not a good measure for grandchild care, as this may be because the grandparents need help from their children, which usually has a negative effect on their children's labor supply (Ettner, 1995; Pavalko & Artis, 1997). If living arrangement is used to measure grandchild care, estimation results will confound the effect of downward (from grandparents to their children) time transfers with that of upward (from adult children to their parents) time transfers. With regard to formal child care, previous studies suggest that the availability of formal child care programs near the household increases the probability of women's LFP (Del Boca & Vuri, 2005; Du & Dong, 2013; Kilburn & Datar, 2002), and formal child care costs usually negatively affect the LFP of mothers of young children (Blau & Robins, 1988; Del Boca & Vuri, 2005; Du & Dong, 2013; Wrohlich, 2004). However, these studies either focus only on formal child care or have aforementioned limitations in measuring informal child care. In view of the limitations of previous studies, this paper contributes to the existing literature in the following aspects. First, the impacts of formal and informal child care on the mother's labor supply are estimated simultaneously. Second, to avoid underestimating the effect of grandchild care on mother labor supply and selection bias, both resident and non-resident grandparents are considered when defining grandchild care. Third, an explicit measure of grandchild care is adopted instead of grandparents' living arrangement to obtain an uncontaminated effect of grandchild care. Last, following Dimova and Wolff (2011), a recursive model with instrumental variables is used to account for potential endogeneity of different forms of child care uses to mother labor supply. As noted by Dimova and Wolff (2011), when formal child care is not available at low costs and the labor market is not flexible, it is difficult for mothers of young children to pursue a career without the help from family members. In China, publicly subsidized child care programs have decreased after the reforms since the middle of the 1990s and become less flexible in accessibility. Private child care programs are more flexible in terms of accessibility and possibly time schedules; however, they are more expensive. Therefore, I expect that formal child care in the form of kindergartens has limited effect on mother labor supply, while informal child care in the form of grandchild care plays an important role. The rest of the paper is organized as follows. Section 2 describes the data, briefly discusses child care policies in China, and gives descriptive statistics. Section 3 presents econometric model specifications and estimation results. Section 4 concludes. 2. Data and descriptive statistics 2.1. Data from the RUMiC surveys I use the data from the Project on Rural–Urban Migration in China (RUMiC) carried out at the beginning of 2008 for the year 4

For a detailed description of child care program reforms in China, see Du and Dong (2013).

228

China Economic Review 44 (2017) 227–240

Y. Li

2007. The survey is designated RUMiC2008 hereafter. This project was initiated by a group of researchers at the Australian National University and Beijing Normal University and was supported by the China National Bureau of Statistics (NBS) and the Institute for the Study of Labor. The questionnaire was designed by Chinese and foreign researchers. Three independent surveys were conducted: the Rural Household Survey, the Urban Household Survey, and the Urban Migrant Survey. The first two were conducted by the NBS using a random sample from the standard annual household income and expenditure surveys that the NBS carries out in cities and rural areas. The Urban Migrant Survey was conducted by the RUMiC project team in collaboration with a professional survey company.5 The data from RUMiC surveys are widely used in existing studies on the labor market in China (Démurger, Li, & Yang, 2012; Frank Qu & Zhao, 2014; Hu & Qian, 2015). I focus on urban residents from the Urban Household Survey, i.e., residents who live in urban areas and hold an urban household registration (so-called urban Hukou) at the time of the interview. I exclude from analysis the rural residents covered by the Rural Household Survey, i.e., farmers living in rural areas and holding a rural household registration. The reason is that urban and rural residents differ in labor choices. As long as the farmers have access to land, they are not regarded as unemployed. If I include both urban residents with urban registrations and rural residents in the estimation sample, the results will be biased. I also exclude from analysis the urban residents without urban registrations, i.e., rural–urban migrants, who are the focus of the Urban Migrant Survey. First, labor choices are different between the urban residents and the migrants. The migrants may choose to either migrate to the urban areas or stay in the rural areas. If those who migrate to the urban areas are ambitious in finding employment in the non-agricultural sector, including migrants in the sample will lead to biased estimates. Second, migrant parents usually leave their children in the countryside with other family members (Duan, LV, Guo, & Wang, 2013). Therefore, whether the child attends kindergarten should not affect the labor supply of the migrant parents. Last, if migrant parents bring their children with them to the urban areas, they are the least privileged in sending their children to public kindergartens or schools. Hence, both labor and child care choices differ between local urban residents and rural-urban migrants. The RUMiC2008 Urban Household Survey is based on a sample of 5002 households from nine province-level administrative units in China (Shanghai, Jiangsu, Zhejiang, Hubei, Chongqing, Guandong, Henan, Anhui and Sichuan) with a total of 14,683 respondents. Considering that the survey contains information on who mostly takes care of the preschooler, I define a benchmark sample that contains mothers who have at least one preschooler. There are 677 out of 5002 households with at least one preschooler at the time of the interview, among which 31 households are dropped, as they are skipped-generation households, i.e., households with no parent of the preschooler. Of the remaining 646 households, 640 have a mother of preschooler(s). For 26 of these 640 mothers, the information on their husbands is missing because the mothers are divorced, separated, or widowed before the interview. Mothers without husbands present in the household can be a special population who are more eager to work and are given priority by the grandparents in taking care of their kids. Including them in the sample will lead to biased estimates. Thus, this paper focuses on mothers of preschoolers whose husbands are present in the household at the time of the interview. To study the relationship among at least three generations, the sample is further restricted to mothers who have at least one living parent or parent-in-law. Only 3 mothers are dropped for having no living parent or parent-in-law. Finally, the benchmark sample contains 611 mothers. For approximately 88% of these mothers, at least two grandparents are still living at the time of the interview, while for approximately 76% of these mothers, at least four grandparents are still living. The RUMiC2008 contains several questionnaires. The household questionnaire provides comprehensive information on demographic and socioeconomic characteristics of each household member, such as age, categorical educational attainment, number of children, employment status, working hours per week, and monthly earnings. The categorical educational attainment takes values from 1 to 4, indicating secondary education, polytechnic college, undergraduate, and postgraduate education. If a household member is employed, he/she can be either employed by others or self-employed. Monthly earnings of an employee consist of wage, subsidies, and bonus both in cash and in kind, while monthly earnings of a business owner is the net income from the business. The education questionnaire focuses on household members who are either under age 16 or at school at the time of the interview. The education questionnaire contains specific information on preschoolers, such as who mostly takes care of the preschooler, whether the child attends kindergarten, monthly kindergarten fees, whether a babysitter is hired for the child, and monthly babysitter fees. The information on school-age children includes school attending status, schooling fees, etc. The parent questionnaire provides demographic and socioeconomic characteristics of the parents and parents-in-law of the household head who are not living with the household head at the time of the interview, including whether they are living, age, gender, categorical educational attainment, etc. The educational attainment takes values from 1 to 3, indicating primary, secondary, and tertiary education. The community questionnaire contains characteristics of the community where the household is located, such as the number of usual residents, the number of the unemployed and laid off, whether the kindergarten that the majority of children living in the community attend has enrolled migrant children, and the number of firms of different types.6 The firm types consist of public firms, private firms (including foreign firms), and self-employed firms. By exploiting the information on the relationship to the household head, I identify the mother and father of the preschooler from

5 The sampling procedure and survey method are described in detail in the RUMiC project survey documentation, see Kong (2010). Additional information about the RUMiC project may be obtained from the China Institute for Income Distribution (website: www.ciidbnu.org/chip/index.asp?lang=EN (last accessed on March 17, 2017)). 6 Communities are administrative units subordinate to districts and counties. Districts and counties are subordinate to cities. Cities are subordinate to provinces. There are 787 communities, 94 districts/counties, and 18 cities covered by the Urban Household Survey of RUMiC2008.

229

China Economic Review 44 (2017) 227–240

Y. Li

the household questionnaire. In addition, I identify the parents and parents-in-law of the mother of the preschooler from both the household questionnaire and the parent questionnaire. Thus, I obtain information on both resident and non-resident grandparents. Because of data limitations, if a preschooler is taken care of mostly by grandparent(s), it is uncertain which of the four potential grandparents is providing care. In other words, I cannot distinguish whether the grandchild care is given by a grandmother or a grandfather, and neither can I distinguish whether the care is given by a paternal or a maternal grandparent. Two types of households are identified. The first type is households with three generations living together, i.e., the preschooler(s), the parents, and the grandparents. The second type is households with just two generations, i.e., the preschooler(s) and the parents. With the benchmark sample, the percentage of households with just two generations is 68%. This household composition characteristic in current China makes it essential to account for non-resident grandparents while studying the effect of grandchild care. 2.2. Variable definitions Now let us describe the main variables of interest. Two dependent variables are examined, i.e., LFP status and labor hour supply of the mother. LFP status is defined as a dummy that takes the value of one if the mother has worked for at least 1 h for pay or has a job but is temporarily on leave during the week prior to the interview, regardless of the job type. Labor hour supply is measured as the average working hours per week, regardless of the job type. The first independent variable of interest is the mother's informal child care use in the form of grandchild care.7 For the benchmark sample, I define grandchild care use as a dummy that takes the value of one if any preschooler of the mother is mostly taken care of by the grandparents, where the grandparents do not necessarily live with the preschooler. The second independent variable of interest is the mother's formal child care use. In urban China, since the child care reforms started in the middle of the 1990s, publicly subsidized child care programs have decreased, while private care centers have increased. As a result, during the period studied in this paper, most public kindergartens only admit children aged 3 years old or above and give priority to children whose parents hold local urban registrations. With the benchmark sample of this paper, of the 342 preschoolers who attend kindergarten, only 5 are younger than 3. In addition, the time schedule of public kindergartens lacks flexibility; it is barely compatible with full-time employment opportunities. Private care centers are more flexible in accessibility and time schedules, but they are more expensive. Du and Dong (2010) note that during the period from 1991 to 1993, grandmothers and formal child care programs appear to be substitutes, while the substitutability between grandmothers and formal child care programs declines from 1997 to 2006. Compared with the period from 1991 to 1993, working mothers have increased their reliance on informal child care to meet their needs from 1997 to 2006. Paid informal child care is similar to private child care centers in terms of flexibility and prices. Because of immature markets and the high prices of babysitters and nannies, only few mothers—13 of 611 of the benchmark sample—employ nannies to look after the preschooler, with half of the preschoolers under age 3 years old. The data show that only households with relatively high earnings can afford it. Given the small usage of nannies, I combine the information on kindergarten and nannies and measure the mother's formal child care use with a binary variable that takes the value of one if any preschooler of the mother attends kindergarten or is looked after by a paid nanny. To avoid estimation bias, I include in regressions the mother's characteristics, namely, age in years, categorical educational attainment, whether having a child under age 2, monthly earnings of the husband, and categorical educational attainment of the mother's parents. Community characteristics are also included, i.e., the number of usual residents of the community, the share of the unemployed and the laid off relative to the number of usual residents, and the number of firms of different types. To control for the size of the firms, based on the information available in the survey, the number of firms of each type is set to be zero if the sum of workers in each type of firm within or surrounding the community is < 100. Apart from the information available in the survey, several variables computed from the sample are also included in regressions, namely, average female monthly earnings of a county/ district and median monthly kindergarten fee at the county/district level. Instrumental variables for the grandchild care and formal child care variable should affect the use of the two types of child care but not directly affect mother labor supply. The first instrument used is the number of children of the grandparents. Considering that a preschooler may have at most four grandparents, I choose the smallest number of children of the four grandparents.8 Previous literature reveals that better-educated grandparents tend to have better-educated and fewer children. Moreover, having fewer children is associated with greater investment in each child, which will impact child labor supply in a number of ways. One advantage of this instrumental variable is that it is not necessarily correlated with mother educational attainments or other human capital characteristics, as the relevant grandparent can be either parent or parent-in-law of the mother. For the consideration that this instrument might be correlated with mother educational attainments or other human capital characteristics, I include as control variables the educational attainments of both the mother's father (maternal grandfather of the preschooler) and the mother's mother (maternal grandmother of the preschooler).9 If there are factors other than grandparent educational attainments influencing mother 7 In this paper, I assume that child care decisions are made by the mother. I assume that the grandparents will provide child care under two conditions, i.e., the mother asks for help and the grandparents are available. Similarly, I assume that the preschooler will attend kindergarten under two conditions, i.e., the mother wants to send the child to kindergarten and the kindergarten is available. 8 Suppose a situation where the paternal grandparents have a large number of children, while the maternal grandparents have a small number of children; the probability of the maternal grandparents providing child care is high, although the probability of the paternal grandparents providing child care is low. 9 I have tried to control for educational attainments of the paternal grandparents, but they show insignificant estimated effects, and the results do not vary much with or without these variables. Therefore, these variables are not included in the final regressions.

230

China Economic Review 44 (2017) 227–240

Y. Li

human capital, the instrument is not purely exogenous and the estimated effects of the two types of child care could be biased. The second instrumental variable used is a certain characteristic of the kindergarten that the majority of the children in the community attend, i.e., whether the kindergarten has enrolled migrant children who do not hold local urban registrations.10 On the one hand, it is reasonable to believe that the kindergarten that the majority of children living in the community attend is likely a public one. On the other, according to current policies in China, public kindergartens in cities give priority to children whose parents hold local urban registrations. If the public kindergarten has enrolled migrant children, it may indicate that the kindergarten supply in the community is sufficient. In the full benchmark sample, there are approximately 65% of the women living in communities with a kindergarten nearby that has enrolled migrant children. It is possible that in more developed communities, there are more migrant workers and higher female labor force participation. Hence, kindergartens in these communities have a higher likelihood to have immigrant children. Nevertheless, after controlling for community economic conditions, this endogeneity issue could presumably be alleviated. The relevance of these instruments will be further discussed in Section 3.

2.3. Descriptive statistics Each observation denotes a mother in the benchmark sample. Sample means and standard deviations from the benchmark sample are reported in Table 1 on characteristics of the grandparents, the kindergarten, the mother, and the father by whether any preschooler of the mother receives grandchild care or formal child care. Looking at columns 1 and 2 of panel A, we observe that grandparents who have fewer children are more likely to provide grandchild care. In contrast, figures in columns 3 and 4 of panel A show that preschoolers whose grandparents have more children are more likely to attend kindergarten. Maternal grandparents who have secondary or tertiary education seem to be more likely to provide grandchild care. There is no clear correlation between formal child care use and the educational attainments of the maternal grandparents. The figures in panel B show that child attending kindergarten is positively associated with the instrumental variable that the kindergarten has enrolled migrant children, whereas grandparents taking care of the child are not clearly associated with this kindergarten feature. This may provide certain evidence that kindergarten attendance is more likely in communities where the kindergarten supply is sufficient. The figures in columns 1 and 2 of panel C show that mothers who receive grandchild care are more likely to participate in labor force and work for longer hours. It seems that mothers who obtained polytech or undergraduate education and mothers who have at least one child under age 2 are more likely to receive grandchild care. Looking at columns 3 and 4, we see that the mother whose child goes to kindergarten is more likely to participate in the labor force and work for longer hours. Mothers with a postgraduate education and mothers who do not have any child under age 2 are more likely to send their children to kindergarten. Panel D gives information on the fathers. There is no clear positive correlation between labor supply of the father and the receiving of grandchild care. This is in accord with previous sociology studies about the situation in Europe (Tobío, 2001). Neither is there an evident positive correlation between father labor supply and the preschooler attending kindergarten. Nevertheless, better educated fathers are likely to receive grandchild care, while fathers with an undergraduate or postgraduate education and fathers with higher earnings are more likely to send their children to kindergartens. The positive correlation between grandchild care and better human capital characteristics of the parents may imply endogeneity. It may be that parents with strong working preferences ask their parents to look after their preschoolers. The same argument applies for the formal child care. Hence, it is essential to control for endogeneity in regressions. Table 2 presents characteristics of the community, features of the mother, and a summary of the instrumental variables by the mother's LFP status. From the figures in panel A, we observe that women in communities with larger populations, lower shares of the unemployed and the laid off, greater numbers of self-employed firms, and lower costs of kindergartens are more likely to participate in the labor force. The negative association between mother LFP and average female earnings of the county/district could indicate that average female earnings and average male earnings are highly correlated and the latter holds women back from the labor market. The figures in panel B show that, again, grandchild care use and kindergarten attendance of the preschooler are positively correlated with the mother's LFP. Better educated mothers and mothers with no child under age 2 are more likely to work. LFP is a decreasing function in market earnings of the spouse and the number of children. Looking at the instruments in panel C, the smallest number of children of the grandparents is negatively associated with mother's LFP, possibly implying that the more children the grandparents have, the less likely it is that the grandparents provide child care and the less likely it is that the mother participates in the labor force. Whether the kindergarten that the majority of children in the community attend has enrolled migrant children is not clearly related with mother's LFP. It could be that even if this instrument is correlated with preschooler's kindergarten attendance, the use of kindergarten does not affect the mother's labor supply. To check whether this is true, we need to run regressions.

10 Because of data limitations, the information on the kindergarten each preschooler attends is not available. The data set only gives information on the kindergarten the majority of the children in the community attend.

231

China Economic Review 44 (2017) 227–240

Y. Li

Table 1 Descriptive statistics by informal and formal child care use. Grandchild care No (1)

Kindergarten Yes (2)

No (3)

Yes (4)

A. Characteristics of the grandparents Number of children of grandparents Maternal grandfather: primary education Maternal grandfather: secondary education Maternal grandfather: tertiary education Maternal grandmother: primary education Maternal grandmother: secondary education Maternal grandmother: tertiary education

2.34 0.47 0.47 0.06 0.45 0.47 0.08

B. Characteristics of the kindergarten Whether the kindergarten enrolls migrant kids

0.66 (0.47)

0.66 (0.48)

0.62 (0.49)

0.69 (0.46)

C. Characteristics of the mother Work status Working hours per week Labor market monthly earnings in yuan Age in years Secondary education Polytech education Undergraduate education Graduate education Having a child aged under 2

0.70 (0.46) 30.5 (22.5) 1468 (2069) 31.8 (4.0) 0.57 (0.50) 0.25 (0.43) 0.15 (0.36) 0.03 (0.17) 0.29 (0.45)

0.94 (0.25) 41.3 (14.1) 2171 (1653) 30.6 (3.9) 0.39 (0.49) 0.31 (0.46) 0.28 (0.45) 0.02 (0.13) 0.45 (0.50)

0.71 (0.46) 30.9 (21.8) 1517 (1667) 30.0 (4.0) 0.51 (0.50) 0.28 (0.45) 0.20 (0.40) 0.008 (0.09) 0.73 (0.44)

0.82 (0.39) 36.0 (19.9) 1812 (2175) 32.5 (3.7) 0.52 (0.50) 0.26 (0.44) 0.18 (0.38) 0.041 (0.20) 0.04 (0.20)

D. Characteristics of the father Work status Working hours per week Labor market monthly earnings in yuan Age in years Secondary education Polytech education Undergraduate education Graduate education Having a child aged under 2 Number of observations

0.97 (0.16) 44.6 (14.8) 2974 (2628) 34.9 (5.3) 0.46 (0.50) 0.26 (0.44) 0.23 (0.42) 0.04 (0.20) 0.29 (0.45) 424

0.95 (0.23) 42.3 (14.5) 2851 (1938) 33.1 (4.4) 0.36 (0.48) 0.30 (0.46) 0.28 (0.45) 0.06 (0.24) 0.45 (0.50) 187

0.95 (0.21) 43.2 (15.4) 2740 (1936) 32.8 (4.7) 0.43 (0.50) 0.31 (0.46) 0.23 (0.42) 0.027 (0.16) 0.73 (0.44) 263

0.97 (0.16) 44.4 (14.2) 3085 (2749) 35.5 (5.1) 0.43 (0.50) 0.24 (0.43) 0.26 (0.44) 0.063 (0.24) 0.04 (0.20) 348

(1.16) (0.50) (0.50) (0.24) (0.50) (0.50) (0.27)

1.80 0.41 0.49 0.11 0.36 0.51 0.13

(0.84) (0.49) (0.50) (0.31) (0.48) (0.50) (0.34)

1.93 0.45 0.48 0.07 0.41 0.48 0.11

(1.01) (0.50) (0.50) (0.26) (0.49) (0.50) (0.31)

2.36 0.45 0.48 0.07 0.43 0.48 0.09

(1.13) (0.50) (0.50) (0.26) (0.50) (0.50) (0.28)

Note: Sample means and standard deviations in parentheses from the benchmark sample of mothers who has at least one preschooler are reported in the table. Figures in columns 1 are from the subsample of mothers who does not receive grandchild care while those in column 2 are from the subsample of mothers who receives grandchild care. Figures in columns 3 are from the subsample of mothers who has no kid attending kindergarten while those in column 4 are from the subsample of mothers who has at least one kid attending kindergarten.

3. Econometric analysis The impacts of formal and informal child care use on the labor supply of mothers of young children are assessed. Both the mother's LFP and labor hour supply are examined. 3.1. Empirical methodology I assume that the mother of young children makes labor decisions and child care decisions simultaneously. In this section, two models that are used to estimate the effects of formal and informal child care on mother labor supply are introduced. Following Dimova and Wolff (2011), a recursive model is employed that consists of an equation for the mother's LFP decision, an equation for the grandchild care decision, and an equation for the formal child care decision.11 To obtain exogenous variations in grandchild care and formal child care uses, I include in the two child care equations instrumental variables that presumably affect grandchild care and formal child care use but do not directly affect mother's LFP. Specifically, I estimate the following model.

L i∗ = αl Gi + βl Ki + γl′Xi + εli

(1)

Gi∗

= γg′Xi + δg′Z i + εgi

(2)

Ki∗ = γk′Xi + δk′Z i + εki

(3)

11 The author is grateful to one anonymous referee for the suggestion that the Stata command “cmp” would permit estimating a model where those observations with only children under 3 do not contribute to the formal child care equation in the likelihood function. Therefore, the full benchmark sample will be used to estimate this recursive model, although children under age 3 usually do not attend kindergartens.

232

China Economic Review 44 (2017) 227–240

Y. Li

Table 2 Descriptive statistics by mother labor force participation status. Mother LFP No (1)

Yes (2)

A. Characteristics of the community Number of usual residents Share of the unemployed or the laid off Average female monthly earnings of the county/district in yuan Number of public firms Number of private firms Number of self-employed firms Median monthly kindergarten fee of the county/district in yuan

8458 (10,539) 6.9% (0.11) 2072 (754) 4.22 (9.59) 16.37 (42.1) 46.11 (106.1) 567 (469)

9044 (9675) 5.4% (0.07) 1946 (610) 4.16 (8.95) 15.38 (41.0) 47.06 (95.0) 549 (444)

B. Characteristics of the mother Grandchild care use Kindergarten use Age in years Secondary education Polytech education Undergraduate education Graduate education Having a child aged under 2 spouse monthly earnings in yuan Number of kids

0.09 (0.28) 0.45 (0.50) 31.8 (4.6) 0.74 (0.44) 0.17 (0.37) 0.08 (0.27) 0.01 (0.12) 0.43 (0.50) 3458 (3072) 1.13 (0.34)

0.37 (0.48) 0.60 (0.49) 31.3 (3.8) 0.45 (0.50) 0.30 (0.46) 0.22 (0.41) 0.03 (0.17) 0.31 (0.46) 2777 (2204) 1.06 (0.27)

C. Instrumental variables Number of children of grandparents Whether the kindergarten enrolls migrant kids Number of observations

2.40 (1.32) 0.67 (0.47) 145

2.10 (1.02) 0.66 (0.47) 466

Note: Sample means and standard deviations in parentheses from the benchmark sample of mothers who has at least one preschooler are reported in the table. Figures in columns 1 are from the subsample of mothers who does not participate in labor force while those in column 2 are from the subsample of mothers who participate in labor force.

where i denotes the mother of the preschooler(s). Li ∗ is a latent variable, indicating the mother's propensity to work. The observed LFP status is denoted by Li such that Li = 1 if Li ∗ > 0 and Li = 0 otherwise. G is a variable indicating whether any preschooler of the mother is mostly taken care of by grandparents. Variable K takes the value of one if any preschooler of the mother attends kindergarten or is looked after by a paid nanny. As robustness checks, in Section 3.4, I will use an alternative estimation sample of mothers who have no preschoolers but have at least one child under age 12. A drawback about this alternative estimation sample is that the survey does not give precise information on who mostly takes care of the child. Hence, I define the variable of grandchild care use to be one if any of the mother's children are living with the grandparents or the non-resident grandparents provide some child care. With this alternative sample, only the effect of informal child care on mother's labor supply is assessed. The vector X is a set of exogenous covariates including the mother's characteristics, namely, age, four educational attainment dummies (each indicates secondary education, polytechnic college, undergraduate, and postgraduate education) with secondary education as the reference category, whether the mother has a child under age 2, spouse monthly earnings in log, and three educational attainment dummies for the maternal grandfather and the maternal grandmother (each indicates primary, secondary, and tertiary education) with primary education as the reference category. Considering that the placement of formal child care programs in a community may suffer endogeneity, i.e., people living in areas where the economy is better developed or with more job opportunities are more willing to join the labor force and require the establishment of child care programs, community characteristics are included in X. Community characteristics include the number of usual residents in log, the share of the unemployed and the laid off relative to the usual residents, average monthly female earnings of the county/district in log, the number of public firms in log, the number of private firms in log, and the number of self-employed firms in log. As public child care policies are quite consistent within the county/district in terms of kindergarten enrollment requirements and fees, the median monthly kindergarten fee at the county/ district level in log and county/district dummies are also included in X to control for policy and economic differences across regions. αl and βl are the coefficients of interest. Gi ∗ and Ki ∗ are the latent variables indicating the mother's propensity to use grandchild care and formal child care, respectively. The observed grandchild care use is denoted by Gi such that Gi = 1 if Gi ∗ > 0 and Gi = 0 otherwise. Similarly, the observed formal care use is denoted by Ki such that Ki = 1 if Ki ∗ > 0 and Ki = 0 otherwise. Vector Z is a set of exogenous covariates that are assumed not to directly affect the mother's LFP, i.e., are uncorrelated with the error term εl. These variables are supposed to provide exogenous variations in the use of the two types of child care and control for endogeneity likely arising from reverse causality between mother labor decision and child care decisions. Specifically, Z contains the number of children of the grandparents. I expect that the more children the grandparent has, the less likely it is that the grandparent provides child care and the more likely it is that the preschooler attends kindergarten. Z also contains a kindergarten characteristic, i.e., whether the kindergarten that the majority of the children of 233

China Economic Review 44 (2017) 227–240

Y. Li

the community attend has enrolled migrant children. I expect that this variable would positively affect the mother's formal child care use and probably negatively affect the mother's grandchild care use. To address the simultaneity of the mother's decision makings, I assume that the error terms εli, εgi, and εki are correlated and follow a standard trivariate normal distribution such that

(εli , εgi , εki ) ∼ N (0, 0, 0; 1, 1, 1, ρlg , ρlk , ρgk )

(4)

where ρlg, ρlk, and ρgk are coefficients of correlation between εli and εgi, εli and εki, and εgi and εki. Eqs. (1), (2), and (3) define a trivariate probit model. For the mother's weekly paid working hours, I use the following model.

Hi = λL i + αh Gi + βh Ki + γh′Xi + εhi

(5)

L i∗ = αl Gi + βl Ki + γl′Xi + εli

(6)

Gi∗ = γg′Xi + δg′Z i + εgi

(7)

Ki∗

(8)

= γk′Xi + δk′Z i + εki

Hi is the number of hours worked per week of the mother. Li is the observed LFP status. Variables G, K, and X are the same as those in Eq. (1). Dependent and independent variables of Eqs. (6), (7), and (8) are defined to be the same as those of Eqs. (1), (2), and (3), respectively. Similarly, I assume the error terms of Eqs. (5), (6), (7), and (8) are correlated and follow a standard multivariate normal distribution such that

(εhi , εli , εgi , εki ) ∼ N (0, 0, 0, 0; 1, 1, 1, 1, ρhl , ρhg , ρhk , ρlg , ρlk , ρgk )

(9)

where ρhl, ρhg, ρhk, ρlg, ρlk, and ρgk are coefficients of correlation between εhi and εli, εhi and εgi, εhi and εki, εli and εgi, εli and εki, and εgi and εki. As noted by Dimova and Wolff (2008), it is usually not easy to estimate a probit model with endogenous discrete variables. Roodman (2011) provides a framework for fitting recursive multiequation systems in which the equation errors are distributed multivariate normal. To estimate the effect of formal and informal child care uses on mother LFP and labor hour supply, I use the simulated likelihood procedure described by Roodman (2011).12 3.2. Effects of informal and formal child care on LFP of mothers of preschoolers Estimation results using the full benchmark sample and applying the model implied by Eqs. (1), (2), and (3) are reported in Table 3. Estimated marginal effects on the probability of the dependent variable being 1 are reported in columns 1, 3, and 5. For continuous regressors, marginal effects are computed at the sample mean; for dummy variables, marginal effects are computed with respect to a change of the variable from 0 to 1.13 Estimated marginal effects in column 1 of Table 3 show that grandchild care has a significant positive impact on mother LFP. If the grandchild care variable changes from 0 to 1, the probability of the mother participating in the labor force will increase by 26.0%. However, the estimated effect of formal child care is positive but not significant. This might be because of admission restrictions of public kindergartens and an inflexible time schedule of the kindergarten that is not compatible with the time schedule of full-time jobs.14 According to a report by the Ministry of Education of China, until 2009, preschool-education, i.e., kindergartens, was the weakest part of all levels of education in China; there was a shortage of education resources in some places.15 This might be another evidence that formal child care does not have significant impact on the mother's labor supply. Estimates of other control variables in column 1 are in line with conventions. The median kindergarten fee at the county/district level has a negative impact on the LFP of the mother, although it is not significant. The insignificance might be because there are two types of kindergartens, i.e., public and private, each having its own price system. The median kindergarten fee at the county/district level does not perfectly reflect the formal child care costs at the household level. The size of the community population has a positive impact, while the share of the unemployed and the laid off has a negative impact on mother LFP, although both effects are 12 Specifically, I use the Stata command “cmp” to estimate the recursive model of the three equations, i.e., Eqs. (1), (2), and (3), and the model of the four equations, i.e., Eqs. (5), (6), (7), and (8). Compared with the method used by Dimova and Wolff (2011), one advantage of the command “cmp” is that it gives marginal effect estimates for probit models and not just coefficient estimates. 13 The Stata command “cmp” stands for conditional mixed process. According to Roodman (2011), “conditional” means that the model can vary by observation. As I include county/district dummies in all regressions, if there is no preschooler being taken care of mostly by grandparents or no preschooler attending kindergarten in a county/district, observations from this county/district will be dropped. To equate the number of observations reported in Table 3 and those reported in Table 1 and Table 2, I group certain counties with few observations into an “other county/district” category by making sure that the grouped counties are in the same city. 14 Because of data limitation, I cannot test directly whether kindergartens that have a flexible time schedule are effective in helping the mother to participate in the labor force. According to certain websites, in 2009, public kindergartens accounted for approximately 77% of all kindergartens in 31 districts/counties of 8 cities that are included in the benchmark sample of this paper (see http://sh.iyaya.com/zhinan/yuanxiao/ (last accessed on March 17, 2017)). Almost all public kindergartens give priority to children whose parents hold local urban registrations and only admit children aged 3years old or above. More than half of these public kindergartens require the children to leave before five o'clock in the afternoon. Private kindergartens do not require local urban registrations of the parents but are more expensive than public kindergartens. 15 See http://www.gov.cn/jrzg/2010-12/03/content_1758599.htm (last accessed on March 17, 2017).

234

China Economic Review 44 (2017) 227–240

Y. Li

Table 3 A trivariate probit model for labor force participation of mothers of preschoolers. LFP

Grandchild care

Kindergarten

MEs (1)

Std. Err. (2)

MEs (3)

Std. Err. (4)

MEs (5)

Std. Err. (6)

A. Community characteristics Log of median kindergarten fee Log of usual residents Share of the unemployed Log of average female earnings Log of public firms Log of private firms Log of self-employed firms

− 0.002 0.038 − 0.137 0.048 0.017 − 0.029* 0.033***

(0.068) (0.025) (0.282) (0.136) (0.023) (0.016) (0.011)

0.004 0.032 −0.498 −0.067 −0.033 0.038** 0.001

(0.077) (0.037) (0.349) (0.139) (0.023) (0.016) (0.012)

− 0.017 − 0.054 − 0.521 0.365* − 0.022 − 0.014 0.021*

(0.115) (0.057) (0.448) (0.212) (0.035) (0.025) (0.012)

B. Mother characteristics Age in years Polytechnic college education Undergraduate education Postgraduate education Having a kid aged under 2 Log of spouse earnings Secondary educ of maternal grandpa Tertiary educ of maternal grandpa Secondary educ of maternal grandma Tertiary educ of maternal grandma

− 0.007 0.121*** 0.076** 0.098 − 0.112 − 0.034* 0.097** 0.064 − 0.074 − 0.064

(0.006) (0.045) (0.039) (0.072) (0.163) (0.019) (0.042) (0.071) (0.045) (0.097)

−0.004 0.127** 0.276*** 0.138 0.140*** −0.005 −0.022 0.005 0.031 0.088

(0.006) (0.059) (0.069) (0.170) (0.051) (0.014) (0.049) (0.088) (0.050) (0.089)

0.009 0.035 0.121 0.359*** − 0.815*** 0.030* − 0.059 − 0.079 0.014 0.039

(0.010) (0.077) (0.084) (0.062) (0.032) (0.016) (0.075) (0.138) (0.075) (0.118)

C. Child care uses Grandchild care Kindergarten

0.260** 0.096

(0.117) (0.183) −0.063*** −0.015 0.032

(0.024) (0.041) 0.023

0.079** 0.137**

(0.038) (0.070)

D. Instrumental variables Number of children of grandparents Kindergarten enrolls migrant kids P-value of test on joint significance of instruments P-value of test on ρlg = 0 P-value of test on ρlk = 0 P-value of test on ρgk = 0 Number of observations

0.031 0.542 0.000 611

Note: Estimated marginal effects of Eqs. (1), (2), and (3) using the benchmark sample of mothers of preschoolers are reported in columns 1, 3, and 5. Standard errors are reported in columns 2, 4, and 6. * means significant at the 10% level, ** means significant at the 5% level, and *** means significant at the 1% level.

insignificant. The number of private firms has a negative and significant impact on mother LFP, whereas the number of self-employed firms has a positive and significant impact. This could indicate that self-employed firms provide job opportunities with a more flexible time schedule than private firms, and therefore the mother is more willing to work. Better educated women are more likely to join the labor force, and spouse earnings have a negative influence on women's LFP. Better educated maternal grandfathers tend to have daughters who are more likely to participate in labor force, whereas the educational attainment of maternal grandmothers has no significant impact on the mother's labor decision. The estimates in column 3 suggest that mothers living in communities with more private firms are more likely to use grandchild care, implying that private firms offer jobs with an inflexible time schedule and mothers are in need of grandchild care if they want to participate in the labor force. Better educated mothers and mothers with a child under age 2 are more likely to be helped by grandparents. Regarding instrumental variables, the number of children the grandparents have negatively affects the probability of grandchild care use, whereas whether the kindergarten has enrolled migrants children does not have a significant impact on grandchild care. To check the relevance of the two instruments, the p-value of the test on the joint significance of the instruments is reported at the bottom of the table. The p-value rejects the null hypothesis that both instruments are equal to zero. The estimates in column 5 indicate that the number of self-employed firms has a positive influence on formal child care use and is significant at the 10% level. This could be evidence that the time schedule of the kindergarten is not compatible with full-time jobs, i.e., the mother who has greater opportunity in self-employed firms has a more flexible schedule to send her children to kindergarten. Women with a postgraduate education are more likely to send their children to kindergartens. Children under age 2 are less likely to attend kindergartens. Spouse earnings positively affect the use of kindergartens. Regarding instrumental variables, the number of children the grandparents have has a significant positive impact on formal child care use. The estimated coefficient of the characteristic of the kindergarten is also significant and has an expected sign. The test on the joint significance of the two instruments rejects the null that both instruments are equal to zero. Finally, p-values of the tests on the null that each of the coefficients of correlation, ρlg, ρlk, and ρgk, is equal to zero are reported at the bottom of Table 3. It is rejected that the coefficients of correlation, ρlg and ρgk, are equal to zero, respectively. As stated before, endogeneity may arise from the fact that mothers with stronger working preferences are more likely to ask their

235

China Economic Review 44 (2017) 227–240

Y. Li

Table 4 A probit and a bivariate probit model for labor force participation of mothers of preschoolers. Probit

Bivariate Probit

LFP

LFP

Grandchild care

MEs (1)

Std. Err. (2)

MEs (3)

Std. Err. (4)

MEs (5)

Std. Err. (6)

A. Community characteristics Log of median kindergarten fee Log of usual residents Share of the unemployed Log of average female earnings Log of public firms Log of private firms Log of self-employed firms

− 0.013 0.040 − 0.206 0.038 0.006 − 0.021 0.032***

(0.063) (0.033) (0.250) (0.117) (0.020) (0.014) (0.011)

− 0.008 0.038 − 0.091 0.062 0.021 − 0.031** 0.031***

(0.068) (0.035) (0.277) (0.128) (0.022) (0.015) (0.011)

−0.012 0.033 −0.473 −0.053 −0.034 0.037** −0.001

(0.078) (0.037) (0.332) (0.141) (0.023) (0.016) (0.012)

B.·Mother characteristics Age in years Polytechnic college education Undergraduate education Postgraduate education Having a kid aged under 2 Log of spouse earnings Secondary educ of maternal grandpa Tertiary educ of maternal grandpa Secondary educ of maternal grandma Tertiary educ of maternal grandma

− 0.010** 0.142*** 0.143*** 0.101** − 0.037 − 0.041*** 0.090** 0.075 − 0.065 − 0.031

(0.005) (0.032) (0.033) (0.050) (0.061) (0.015) (0.040) (0.057) (0.042) (0.087)

− 0.007 0.109*** 0.073** 0.089 − 0.086 − 0.030* 0.097** 0.051 − 0.077* − 0.071

(0.005) (0.044) (0.033) (0.080) (0.066) (0.017) (0.043) (0.075) (0.045) (0.093)

−0.004 0.130** 0.280*** 0.122 0.140*** −0.004 0.026 0.010 0.027 0.079

(0.006) (0.059) (0.069) (0.178) (0.052) (0.014) (0.049) (0.087) (0.050) (0.089)

C. Child care uses Grandchild care Kindergarten

0.320*** 0.142**

(0.053) (0.061)

0.305*** 0.140**

(0.057) (0.058) −0.065*** −0.016 0.021

(0.023) (0.042)

D. Instrumental variables Number of children of grandparents Kindergarten enrolls migrant kids P-value of test on joint significance of instruments P-value of test on ρlg = 0 Number of observations

0.053 611

611

Note: Estimated marginal effects of Eq. (1) using the benchmark sample and applying a simple probit model are reported in columns 1. Estimated marginal effects of Eqs. (1) and (2) using the benchmark sample and applying a Bivariate probit model are reported in columns 3 and 5. Standard errors are reported in columns 2, 4, and 6. * means significant at the 10% level, ** means significant at the 5% level, and *** means significant at the 1% level.

parents or parents-in-law for help and send their children to kindergarten. If this is true, without addressing the endogeneity, the effects of informal and formal child care on mother labor supply can be overestimated. Taking the LFP of mothers of preschoolers for example, a simple probit model implied by Eq. (1) is estimated without instrumenting either the informal or the formal child care use. The results are reported in columns 1 and 2 of Table 4. Additionally, a bivariate probit model implied by Eqs. (1) and (2) is estimated only instrumenting the informal child care use. The results are reported in columns 3 to 6 of Table 4. The estimates in columns 1 and 2 of Table 4 indicate that the estimated effects of both the formal and informal child care are larger than those reported in columns 1 and 2 of Table 3 and the effect of kindergarten attendance becomes significant. The estimates in columns 3 and 4 of Table 4 show that the estimated effects of both the formal and informal child care are again larger than those reported in Table 3 and the effect of kindergarten becomes significant. The enlarged effect of informal child care may be because the two types of child care uses are correlated. The correlation between informal and formal child care uses is verified by the test on the significance of error term correlations between informal and formal child care equations shown at the bottom of Table 3. Estimation results from these two additional models reflect that without controlling for endogeneity, there is risk of overestimating. The instruments adopted in this paper help to address the endogeneity issue. 3.3. Effects of informal and formal child care on labor hours of mothers of preschoolers Estimation results using the full benchmark sample and applying the model implied by Eqs. (5), (6), (7), and (8) are presented in Table 5. For Eq. (5), estimated coefficients are reported in column 1; for Eqs. (6), (7), and (8), estimated marginal effects on the probability of the dependent variable being 1 are reported in columns 3, 5, and 7. Estimated marginal effects for the LFP equation in column 3 of Table 5 are similar to those reported in column 1 of Table 3 in terms of magnitude, sign, and significance. The same is true for the estimates in column 5 of Table 5 compared with those in column 3 of Table 3 and for the estimates in column 7 of Table 5 compared with those in column 5 of Table 3. As shown in column 3, the estimated effect of grandchild care shows that if the grandchild care variable changes from 0 to 1, the probability of the mother 236

China Economic Review 44 (2017) 227–240

Y. Li

Table 5 Labor hour supply conditional on labor force participation of mothers of preschoolers. Hours

LFP

Grandchild care

Kindergarten

Coef. (1)

Std. Err. (2)

MEs (3)

Std. Err. (4)

MEs (5)

Std. Err. (6)

MEs (7)

Std. Err. (8)

A. Community characteristics LFP Log of median kindergarten fee Log of usual residents Share of the unemployed Log of average female earnings Log of public firms Log of private firms Log of self-employed firms

48.15*** − 0.487 − 0.757 − 6.111 − 4.867** − 0.057 − 0.014 0.027

(2.205) (1.358) (0.678) (5.627) (2.367) (0.417) (0.293) (0.215)

− 0.002 0.036 − 0.093 0.114 0.014 − 0.031* 0.027**

(0.075) (0.038) (0.323) (0.142) (0.025) (0.017) (0.012)

− 0.001 0.021 − 0.489 − 0.086 − 0.035 0.034** − 0.001

(0.088) (0.047) (0.346) (0.162) (0.023) (0.016) (0.012)

− 0.010 − 0.074 − 0.554 0.348* − 0.015 − 0.014 0.021*

(0.113) (0.056) (0.449) (0.211) (0.035) (0.024) (0.011)

B.·Mother characteristics Age in years Polytechnic college education Undergraduate education Postgraduate education Having a kid aged under 2 Log of spouse earnings Secondary educ of maternal grandpa Tertiary educ of maternal grandpa Secondary educ of maternal grandma Tertiary educ of maternal grandma

− 0.016 − 4.519*** − 4.253*** − 2.778 − 1.017 − 0.132 1.556* 1.346 − 1.439* − 2.013

(0.114) (1.014) (1.337) (2.241) (2.918) (0.278) (0.857) (1.578) (0.881) (1.477)

− 0.010* 0.118** 0.057* 0.066 − 0.068 − 0.034* 0.094** 0.068 − 0.080 − 0.064

(0.006) (0.055) (0.032) (0.106) (0.176) (0.019) (0.047) (0.088) (0.048) (0.096)

− 0.004 0.127** 0.282*** 0.098 0.140*** − 0.006 − 0.010 0.050 0.031 0.076

(0.007) (0.059) (0.069) (0.173) (0.053) (0.016) (0.051) (0.091) (0.047) (0.084)

0.006 0.042 0.127 0.364*** − 0.816*** 0.033* − 0.037 − 0.046 0.013 0.038

(0.010) (0.076) (0.085) (0.064) (0.032) (0.019) (0.070) (0.127) (0.072) (0.115)

C. Child care uses Grandchild care Kindergarten

− 3.810 − 1.354

(3.363) (3.768)

0.261** 0.103

(0.113) (0.176) − 0.074*** − 0.015 0.019

(0.028) (0.042)

0.086** 0.140** 0.007

(0.036) (0.070)

D. Instrumental variables Number of children of grandparents Kindergarten enrolls migrant kids P-value of test on joint significance of instruments P-value of test on ρhl = 0 P-value of test on ρhg = 0 P-value of test on ρhk = 0 P-value of test on ρlg = 0 P-value of test on ρlk = 0 P-value of test on ρgk = 0 Number of observations

0.084 0.382 0.991 0.024 0.851 0.000 611

Note: Estimation results of Eqs. (5) to (8) using the benchmark sample of mothers of preschoolers are reported. Estimated coefficients of Eq. (5) are reported in column 1. Estimated marginal effects of Eqs. (6), (7), and (8) are reported in columns 3, 5, and 7, respectively. Standard errors are reported in columns 2, 4, 6, and 8. * means significant at the 10% level, ** means significant at the 5% level, and *** means significant at the 1% level.

participating in the labor force will increase by 26.1%. The estimated effect of kindergarten on mother LFP remains insignificant. The estimated coefficients in column 1 show that, after conditioning labor hours on LFP, neither grandchild care nor kindergarten has any significant impact on mothers' labor hours. The estimated effects of both grandchild care and kindergarten are negative and insignificant. Regarding other controls, women's average monthly earnings of the county/district have a significant negative impact on mother labor hour supply, reflecting an income effect. Average working hours per week of the working mothers in the benchmark sample is 42.8; if the average earnings increase by 1%, labor hours per week of the working mother will decrease by 4.8. Better educated mothers tend to work for fewer hours, and mothers whose fathers have secondary education tend to work for more hours once entering the labor market. Finally, p-values of the tests on the null hypotheses that each of the coefficients of correlation, ρhl, ρhg, ρhk, ρlg, ρlk, and ρgk, is equal to zero are reported at the bottom of Table 5. It is rejected that ρhl, ρlg and ρgk are equal to zero, respectively.

3.4. Effects of informal child care on labor supply of mothers of children under 12 As mentioned in Section 3.1, for robustness checks, I use an alternative estimation sample consisting of mothers who have no preschoolers but have at least one child under age 12 who is usually a primary school student. Under the current public policy in China, primary school is a part of compulsory education. The data are in line with the public policy, i.e., all non-preschoolers who are under 12 years old are at school at the time of the interview. Thus, I estimate only the effect of grandchild care on mother LFP and labor hours applying the two recursive models presented in Section 3.1. Estimation results for mother LFP are reported in Table 6 and those for mother labor hours are reported in Table 7. The estimated marginal effects in column 1 show that grandchild care has a significant positive impact on the LFP of the mother. If 237

China Economic Review 44 (2017) 227–240

Y. Li

Table 6 A bivariate Probit model for labor force participation of mothers of primary school kids. LFP

Grandchild care

MEs (1)

Std. Err. (2)

MEs (3)

Std. Err. (4)

A. Community characteristics Log of school fees Log of usual residents Share of the unemployed Log of average female earnings Log of public firms Log of private firms Log of self-employed firms

−0.011 0.035 −0.305* 0.164* −0.014 0.026** 0.012

(0.011) (0.026) (0.164) (0.097) (0.016) (0.012) (0.009)

0.018 − 0.019 − 0.753* − 0.142 − 0.023 0.024 − 0.007

(0.020) (0.045) (0.403) (0.167) (0.028) (0.020) (0.014)

B. Mother characteristics Age in years Age squared Polytechnic college education Undergraduate education Postgraduate education Log of spouse earnings Secondary educ of maternal grandpa Tertiary educ of maternal grandpa Secondary educ of maternal grandma Tertiary educ of maternal grandma

−0.016 0.014 0.173*** 0.157*** 0.058 −0.016 0.062* 0.050 0.024 −0.008

(0.050) (0.064) (0.027) (0.021) (0.102) (0.009) (0.038) (0.094) (0.034) (0.047)

− 0.022 0.034 0.124*** 0.098 − 0.201 0.026 0.028 0.121 0.161*** 0.215**

(0.081) (0.105) (0.062) (0.088) (0.166) (0.019) (0.061) (0.109) (0.058) (0.096)

C. Child care uses Grandchild care

0.180**

(0.083)

D. Instrumental variables Number of children of grandparents P-value of test on ρlg = 0 Number of observations

− 0.067***

(0.024)

0.085 576

Note: Estimated marginal effects of Eqs. (1) and (2) using the sample of mothers of primary school kids are reported in columns 1 and 3. Standard errors are reported in columns 2 and 4. * means significant at the 10% level, ** means significant at the 5% level, and *** means significant at the 1% level.

the grandchild care variable changes from 0 to 1, the probability of the mother participating in labor force will increase by 18.0%. The magnitude of this effect is smaller than that obtained from the benchmark sample shown in Table 3. It is reasonable to think that children become less dependent on child care once they grow older. Thus, the mother is less dependent on grandchild care if she wants to participate in the labor force. As for other controls, the share of the unemployed and the laid off has a significant negative impact on mother LFP. If the share increases by 1%, the probability of the mother participating in the labor force will decrease by 30.5%. Better educated mothers and mothers whose fathers have secondary education are more likely to participate in the labor market. The estimates in column 3 of Table 7 are quite similar to those in column 1 of Table 6 in terms of magnitude, sign, and significance. The estimates in column 5 of Table 7 are quite similar to those in column 3 of Table 6 in terms of magnitude, sign, and significance. As shown in column 3, grandchild care has a significant positive impact on mother LFP. If the grandchild care varies from 0 to 1, the probability of the mother participating in labor force will increase by 18.5%. In both Table 6 and Table 7, the estimated effect of the instrumental variable is highly significant and has an expected sign. The estimated coefficients in column 1 of Table 7 show that, after conditioning working hours on LFP, grandchild care does not have any significant impact on working hours of the mother. Regarding other controls, the average female monthly earnings of the county/district negatively affect the mother's labor hour supply, indicating an income effect. Better educated mothers tend to work fewer hours upon entering the labor market. Spouse earnings negatively influence the labor hour supply of the working mothers. 4. Conclusions In urban China, since the launch of a program of radical restructuring of state-owned enterprises in 1997, publicly subsidized child care programs has decreased. After the reforms, public child care programs have been usually available for elder children and given priority to children whose parents hold local urban registrations. Meanwhile, private child care centers have increased in number, offering the services at higher prices. The reforms might have weakened the effect of formal child care on the labor supply of mothers of young children while increasing the reliance of working mothers on informal child care. In this paper, I study the effects of both informal child care in the form of grandchild care and formal child care in the form of kindergarten or paid nannies on the labor supply of mothers of young children after the reforms. To overcome data and methodology limitations of previous studies about urban China, I employ a recursive model that includes three equations, one for mother labor force participation, one for mother informal child care decision, and another for mother formal child care decision. To study the labor hour supply of the mother, another recursive model is constructed with an additional equation 238

China Economic Review 44 (2017) 227–240

Y. Li

Table 7 Labor hour supply conditional on labor force participation of mothers of primary school kids. Hours

LFP

Grandchild care

Coef. (1)

Std. Err. (2)

MEs (3)

Std. Err. (4)

MEs (5)

Std. Err. (6)

A. Community characteristics LFP Log of school fees Log of usual residents Share of the unemployed Log of average female earnings Log of public firms Log of private firms Log of self-employed firms

46.97*** 0.261 −0.395 −1.321 −5.680** −0.405 −0.094 0.353*

(3.081) (0.300) (0.692) (6.082) (2.615) (0.417) (0.322) (0.215)

−0.017 0.038 −0.283* 0.247** −0.019 0.041*** 0.012

(0.014) (0.033) (0.150) (0.120) (0.020) (0.014) (0.010)

0.016 − 0.011 − 0.817** − 0.227 − 0.027 0.030* − 0.008

(0.021) (0.045) (0.375) (0.163) (0.028) (0.018) (0.013)

B. Mother characteristics Age in years Age squared Polytechnic college education Undergraduate education Postgraduate education Log of spouse earnings Secondary educ of maternal grandpa Tertiary educ of maternal grandpa Secondary educ of maternal grandma Tertiary educ of maternal grandma

−1.941 2.218 −3.706*** −4.235*** 0.097 −0.397* 1.604* 2.569 0.830 −1.063

(1.258) (1.637) (1.130) (1.552) (2.965) (0.242) (0.958) (1.771) (0.962) (1.660)

−0.021 0.017 0.158*** 0.180*** 0.055 −0.018 0.080* 0.065 0.014 −0.006

(0.060) (0.077) (0.043) (0.036) (0.111) (0.011) (0.047) (0.117) (0.058) (0.097)

− 0.015 0.025 0.122** 0.148* − 0.205 0.015 0.029 0.145 0.166*** 0.211**

(0.074) (0.096) (0.062) (0.085) (0.157) (0.016) (0.060) (0.102) (0.054) (0.093)

C. Child care uses Grandchild care

−3.837

(2.425)

0.185**

(0.096)

D. Instrumental variables Number of children of grandparents P-value of test on ρhl = 0 P-value of test on ρhg = 0 P-value of test on ρlg = 0 Number of observations

− 0.076***

(0.024)

0.046 0.055 0.092 576

Note: Estimation results of Eqs. (5) to (7) using the sample of mothers of primary school kids are reported. Estimated coefficients of Eq. (5) are reported in columns 1. Estimated marginal effects of Eqs. (6) and (7) are reported in columns 3 and 5, respectively. Standard errors are reported in columns 2, 4, and 6. * means significant at the 10% level, ** means significant at the 5% level, and *** means significant at the 1% level.

in the aforementioned recursive model that conditions mother labor hour supply on labor force participation. To address potential endogeneity arising from simultaneity or/and reverse causality, for both recursive models, I include certain instrumental variables in the two child care equations. Considering that the mother makes these decisions simultaneously, for each recursive model, I allow for correlations among the error terms of the equations. With a benchmark sample of mothers who have at least one preschooler, I find a significant positive impact of grandchild care on the labor force participation of the mother. Quantitatively, grandchild care use would increase the probability of a mother participating in the labor force by approximately 26%. In contrast, the estimated effect of formal child care on mother labor force participation is positive but not significant. This might be because public kindergartens have strict admission requirements and offer services with a time schedule that is not flexible enough to be compatible with full-time jobs. For the mother's labor hour supply, after conditioning on labor force participation, neither grandchild care nor kindergarten use has a significant impact. Currently, in urban China, there is a tendency to postpone the retiring age. This might reduce the provision of grandchild care and affect the labor supply of mothers of young children. Moreover, if the care provided by grandmothers is not substituted by formal care of equal or better quality, the reduction in grandchild care may have implications on child development. To keep mothers of young children on the labor market and to maintain good health of their children, one possible way could be to provide child care programs with fewer admission requirements, a more flexible time schedule, affordable prices, and reasonable quality. References Blau, D. M., & Robins, P. K. (1988). Child-care costs and family labor supply. The Review of Economics and Statistics, 374–381. Chen, F., Short, S. E., & Entwisle, B. (2000). The impact of grandparental proximity on maternal childcare in China. Population Research and Policy Review, 19(6), 571–590. Connelly, R., Degraff, D. S., & Levison, D. (1996). Women's employment and child care in Brazil. Economic development and cultural change, 619–656. Del Boca, D., & Vuri, D. (2005). Labor supply and child care costs: The effect of rationing. IZA Discussion Paper. Démurger, S., Li, S., & Yang, J. (2012). Earnings differentials between the public and private sectors in China: Exploring changes for urban local residents in the 2000s. China Economic Review, 23(1), 138–153. Dimova, R., & Wolff, F.-C. (2008). Grandchild care transfers by ageing immigrants in France: Intra-household allocation and labour market implications. European Journal of Population/Revue européenne de Démographie, 24(3), 315–340.

239

China Economic Review 44 (2017) 227–240

Y. Li

Dimova, R., & Wolff, F.-C. (2011). Do downward private transfers enhance maternal labor supply? Evidence from around Europe. Journal of Population Economics, 24(3), 911–933. Du, F., & Dong, X.-Y. (2010). Women's labor force participation and childcare choices in urban China during the economic transition. 2010, The University of Winnipeg, Department of Economics World Paper Number4. Du, F., & Dong, X.-Y. (2013). Women's employment and child care choices in urban China during the economic transition. Economic Development and Cultural Change, 62(1), 131–155. Duan, C.-R., LV, L.-D., Guo, J., & Wang, A. Z.-P. (2013). Survival and development of left-behind children in rural China: Based on the analysis of sixth census data. Population Journal, 35, 37–49. Ettner, S. L. (1995). The impact of “parent care” on female labor supply decisions. Demography, 32(1), 63–80. Frank Qu, Z., & Zhao, Z. (2014). Evolution of the Chinese rural-urban migrant labor market from 2002 to 2007. China Agricultural Economic Review, 6(2), 316–334. Hu, A., & Qian, Z. (2015). Educational homogamy and earnings inequality of married couples: Urban China, 1988–2007. Research in Social Stratification and Mobility, 40, 1–15. Kilburn, M. R., & Datar, A. (2002). The availability of child care centers in China and its impact on child care and maternal work decisions. RAND. Kong, S. T. (2010). Rural–urban migration in China: survey design and implementation. Cheltenham, UK: Edward Elgar Publishing, Inc. Ogawa, N., & Ermisch, J. F. (1996). Family structure, home time demands, and the employment patterns of Japanese married women. Journal of Labor Economics, 677–702. Pavalko, E. K., & Artis, J. E. (1997). Women's caregiving and paid work: Causal relationships in late midlife. The Journals of Gerontology Series B: Psychological Sciences and Social Sciences, 52(4), S170–S179. Qu, J.-Y., & Sun, L.-J. (2011). Living arrangements and changes of the elderly in China: From 2000 to 20006. Population Journal, 2, 40–45. Roodman, D. (2011). Fitting fully observed recursive mixed-process models with cmp. Stata Journal, 11(2), 159–206 (48). Sasaki, M. (2002). The causal effect of family structure on labor force participation among Japanese married women. Journal of Human Resources, 429–440. Shen, K., Zhang, Y., & Yan, P. (2012). Family structure and female labor force participation in China. Population Research, 36(5), 15–27. Tobío, C. (2001). En Espagne, la abuela au secours des mères actives. Autrement. Série mutations, 210, 102–115. Wrohlich, K. (2004). Child care costs and mothers' labor supply: An empirical analysis for Germany. Discussion paper, DIW-Diskussionspapiere. Yao, X.-G., & Tan, L. (2005). Family income and labor force participation of married women in urban China. Economic Research Journal, 7, 18–27. Yunrong Li The author obtained her PhD degree in Economics from Universidad Carlos III de Madrid (Spain) in 2014. The author has been working at the School of Public Finance and Taxation at Southwestern University of Finance and Economics since 2014 to the present. The research interests of the author lie in Labor Economics, Development Economics, and issues in China. The author has published papers on the journals of World Development, PLoS One, and Journal of Informetrics.

240