Paid Maternity Leave and Breastfeeding in Urban China

ABSTRACT

Using data from the 2010 Survey on Chinese Women's Social Status, this contribution estimates the effect of paid maternity leave on breastfeeding duration in urban China during the 1988–2008 period. The analysis applies a policy-based identification strategy to control for the endogenous relationship between paid leave entitlements and breastfeeding decisions. Estimates show that paid maternity leave has a strong positive effect on breastfeeding duration. Specifically, if the length of paid leave increases by thirty days, then the probability of breastfeeding for at least six months increases by 12 percentage points. Between 1988 and 2008, the average length of paid leave for mothers without a college education decreased by twenty-three days, which reduced these mothers’ probability of breastfeeding for at least six months by 9 percentage points. These results support the view that paid maternity leave enhances the ability of employed women to sustain breastfeeding and call for universal paid leave entitlements.

KEYWORDS:

• Breastfeeding
• maternal employment
• paid maternity leave
• Chinese women

JEL Codes:

• J01
• J08
• J33

INTRODUCTION

Breastfeeding has many health benefits for infants and mothers. It reduces the risks of many adverse health outcomes, such as sudden infant death syndrome, asthma, diabetes, and obesity, and it enhances an infant's cognitive development and neurodevelopment (Anderson, Johnstone, and Remley 1999; Vestergaard et al. 1999; Ip et al. 2007). Breastfeeding also benefits mothers by reducing the incidence of breast and ovarian cancers (US Department of Health and Human Services 2011). In recognition of these benefits, the World Health Organization (WHO) and the United Nations Children's Fund (UNICEF) recommend exclusive breastfeeding for the first six months of life, followed by continued breastfeeding with complementary foods throughout the rest of the first year and beyond (WHO 2003).

Maintenance of breastfeeding for the optimal duration has proved difficult for mothers who participate in the labor force. Because of its intensity and timing, breastfeeding is difficult to schedule without interfering with wage employment. Researchers have found that mothers who plan to work full time postpartum are less likely to initiate breastfeeding than mothers who expect not to work or who expect to work part time (Lindberg 1996; Fein and Roe 1998; Skafida 2012). Researchers have also found that mothers who return to paid work within a year of delivery are more likely to stop breastfeeding than mothers who return later (Visness and Kennedy 1997; Roe et al. 1999; Guendelman et al. 2009). Early return to paid employment post-birth is widely recognized as a main reason for noninitiation and early cessation of breastfeeding (Dennis 2002).

Paid maternity leave provision is a major policy measure that is intended to make maternal employment more compatible with breastfeeding. By safeguarding women's employment and income security during pregnancy and after childbirth, paid maternity leave reduces the opportunity cost of time spent on breastfeeding and consequently increases the likelihood of optimal breastfeeding duration. In Convention No. 183 adopted in 2000, the International Labour Organization (ILO) called upon member nations to provide for at least fourteen weeks of paid maternity leave at a minimum rate of two-thirds of previous earnings through social insurance or public funds. The ILO further recommended that paid maternity leave be extended to at least eighteen weeks. As of 2014, ninety-eight countries had adopted policies providing maternity leave lasting at least fourteen weeks, although only fifty-seven countries had introduced regulations that fully met the three requirements of Convention No. 183 in terms of leave duration, level of payment, and source of funding (Addati, Cassirer, and Gilchrist 2014).

A small number of studies have empirically investigated the effectiveness of paid maternity leave as a means to promote breastfeeding. Michael Baker and Kevin Milligan (2008) estimated the effect of the expansion of paid maternity leave from twenty-five to fifty weeks on breastfeeding duration in Canada. This expansion was enacted on December 31, 2000. The authors found that the leave expansion increased the length of breastfeeding by one month among mothers eligible for a longer leave and increased the rate of exclusive breastfeeding through the first six months by 7.7 to 9.1 percentage points. Rui Huang and Muzhe Yang (2015) investigated the impacts of the implementation of a paid family leave program in California. The program, enacted on September 23, 2002, has been utilized primarily as maternity leave for mothers to bond with their infants. The authors found that the program implementation increased rates of exclusive breastfeeding for the first three and six months by 3 to 5 percentage points and increased breastfeeding rates for the first three, six, and nine months by 10 to 20 percentage points. The findings generated by Baker and Milligan (2008) and Huang and Yang (2015) provide valuable inputs for the formulation of maternity leave policy, as both studies seek to estimate the causal effects of paid maternity leave on breastfeeding using a policy-based identification strategy.¹

However, much of the research has focused on developed countries, and relatively little is known about how women's breastfeeding decisions are affected by maternal employment and paid maternity leave in developing and transition countries. To fill this knowledge gap, the present study empirically analyzes the impacts of paid maternity leave duration on breastfeeding in urban China. The analysis utilizes data from a nationally representative survey – the third wave of the Survey on Chinese Women's Social Status (SCWSS) conducted in 2010.

As a developing country that has undergone the transition from a centrally planned economy to a market economy since the late 1970s, China provides a unique context for studying the link between women's breastfeeding decisions and labor market policies. Similarly to the ex-socialist countries of Central and Eastern Europe and the Soviet Union, China achieved high women's labor force participation by providing employed mothers with paid maternity leave and subsidized nurseries and kindergartens during the Maoist era (1949–76). The economic transition has eroded the state and employer support for women's reproductive roles, which has led to a dramatic decline in the coverage of paid maternity leave and publicly funded childcare programs, with women of low socioeconomic status being the hardest hit (Liu, Zhang, and Li 2009). The decline of public support has intensified the work–family conflict that Chinese women face. The changing role of the state in care provision and its implications for women's opportunities and gender equality have attracted wide attention from research communities (Cook and Dong 2011). Recent studies have investigated the impacts of changing care arrangements for children and the elderly on women's employment and earnings.² The present study adds to the growing body of literature on work–family conflict in post-reform China by analyzing this conflict from a new angle.

As in Baker and Milligan (2008) and Huang and Yang (2015), our study seeks to control for the potentially endogenous relationship between paid maternity leave entitlements and breastfeeding decisions with a policy-based identification strategy. This strategy explores the variations in maternity leave policy across Chinese provinces. Our analysis shows that paid maternity leave has a positive effect on breastfeeding duration, and that this effect is especially strong for less educated mothers. Specifically, a thirty-day extension of paid maternity leave increases the probability of breastfeeding for at least six months by 11.7 percentage points. We also find that following public-sector restructuring in the late 1990s, the average length of paid leave for mothers without a college education decreased by twenty-three days, which decreased the probability that these mothers breastfed for at least six months by 9 percentage points. These results support the claim that paid maternity leave is an effective means of enhancing employed mothers’ inclination to sustain breastfeeding and call for policy measures to make paid leave accessible to all employed mothers.

POLICY BACKGROUND

During the Maoist era, women's full participation in the labor force played a key role in the Chinese government's efforts to advance women's position in society (Croll 1983). To facilitate women's labor force participation, the Chinese government initiated a series of measures to mitigate the work–family conflict that women face during pregnancy, childbirth, and breastfeeding. These measures included an entitlement to fifty-six days of paid maternity leave, on-site nurseries and breastfeeding rooms in the workplace, and the provision of extra break time to mothers to breastfeed during the workday (Liu, Zhang, and Li 2009). However, under China's household registration system, these generous maternal benefits were granted only to women in the urban sector, with the employees of government agencies, nonprofit public organizations (schools, universities, and hospitals), state-owned enterprises (SOEs), and urban collectives being the main beneficiaries.

In the late 1970s, China embarked on the transition to a market-oriented economy with a gradualist approach. In the first fifteen years of reforms, the Chinese government sought reform within the socialist system, directing its efforts primarily at improving the incentives of state workers and managers and encouraging the development of non-state sectors. Radical reform occurred in the late 1990s when the central government launched a large-scale labor retrenchment program intended to revitalize the ailing SOE sector. This public-sector restructuring has fundamentally changed the landscape of the urban labor market, reducing the public-sector share of employment from 75.7 percent in 1995 to 33.4 percent in 2002 (Dong and Xu 2009). This sharp decline in public sector employment severely weakened the mechanisms through which urban women were protected prior to the reforms.

In the post-reform era, a series of new regulations have been introduced to safeguard women's employment and reproductive rights in the new market economy. The Labor Law adopted in 1995 stipulated that women and men have equal employment rights and that employers shall not lay off women employees or lower their wages for reasons of marriage, pregnancy, maternity leave, or breastfeeding. The Labor Contract Law enacted in 2008 introduced the provision that prohibits employers from unilaterally terminating labor contracts with women employees who are pregnant, give birth, and care for a baby postpartum. Thus, under the Labor Law and Labor Contract Law, women employees are entitled to job-protected maternity leave.

The post-reform era also witnessed a series of improvement in maternity benefits. The length of paid maternity leave was extended from fifty-six days prior to reform, to ninety days in 1988, and to ninety-eight days in 2012.³ Paid maternity leave duration was further extended to a minimum of 128 days in 2016 after the long-standing one-child policy was replaced with a policy that encourages each couple to have two children. The latest extension of paid leave was intended to boost fertility and slow the population aging process. In addition to paid leave, maternity benefits also include a reimbursement of expenses for prenatal examinations, delivery, hospital care, and medicine (Liu, Zhang, and Li 2009). Either maternity insurance or employers pay maternity benefits. The maternity insurance program was first introduced in Shanghai in 2005 and gradually adopted in other provinces. Maternity insurance premiums are paid by the employer based on the size of a company's payroll. For women employees who are covered by maternity insurance, maternity allowance is equal to the company or organization's average monthly wage from the previous year, whereas for a woman employee who is not covered by maternity insurance, the maternity allowance is equal to her basic wage before pregnancy.⁴

Because provincial governments administer maternity benefits, these benefits vary widely across provinces. Since the late 1980s, many provincial governments have used paid maternity leave as an incentive device to promote family planning. Except for Zhejiang and Tibet, all provinces have adopted a policy that grants employed women who had their first birth at age 24 or older extra days of paid maternity leave, in addition to the standard length of ninety days; this policy is intended as a bonus for late childbirth. The policy was enacted in each province at different points in time between 1987 and 2003, and the number of bonus days varied across provinces, ranging from fifteen to ninety days (see Table 1). In some localities, the length of paid maternity leave was further extended for mothers who had abnormal deliveries or reduced as a punishment for one-child policy violations.

Table 1 Paid maternity leave with a bonus for late childbirth by province in urban China

Province	Bonus days of paid maternity leave for late childbirth + 90 days	Time of implementation
Beijing	120	June 1991
Tianjin	120	January 1989
Hebei	135	March 1989
Shanxi	120	October 1999
Inner Mongolia	120	November 1995
Liaoning	150	May 1988
Jilin	104	July 1988
	120	October 1993
Heilongjiang	180	February 1990
Shanghai	105	January 1990
	120	December 2003
Jiangsu	120	October 1990
Zhejiang	90
Anhui	120	December 1988
Fujian	180	July 1988
Jiangxi	120	September 1990
Shandong	150	March 1988
Henan	180	July 1990
Hubei	105	March 1988
	120	March 1997
Hunan	120	January 1990
Guangdong	105	December 1992
Guangxi	104	January 1989
Hainan	105	October 1989
Chongqing	110	July 1987
	120	October 2002
Sichuan	110	July 1987
	120	October 2002
Guizhou	120	July 1998
Yunnan	120	September 2002
Tibet	90
Shaanxi	105	March 1991
Gansu	100	January 1990
Qinghai	105	April 1992
	120	January 2003
Ningxia	104	January 1991
Xinjiang	120	January 2003

Sources: “Family Planning Regulations” and “Regulations on Population and Family Planning” issued by each province.

The implementation of maternity leave has encountered many obstacles in the post-reform period. The maternity leave provision does not benefit all mothers employed in the wage sector, as the maternity leave regulations require. As a reflection of the persistent rural–urban divide, paid maternity leave remains largely an urban phenomenon. In the cities, while permanent employees in the state sector are entitled to maternity leave benefits, many state-sector employees on short-term contracts do not have such benefits. The coverage of paid leave in private sectors is much lower, as there are no effective means of enforcing labor market regulations in private sectors (Liu, Zhang, and Li 2009). In 2012, only 33.5 percent of women employees in cities had maternity insurance (Wang 2014). Under pressure to make profits, enterprises are reluctant to accommodate women's reproductive needs. Enterprise surveys from the early 2000s show that only 5 percent of enterprises provided a breastfeeding room for women employees and that less than 6 percent of enterprises had nurseries and kindergartens in the early 2000s.⁵ Hence, mothers have found it increasingly difficult to combine breastfeeding with employment in the post public-sector restructuring period. The work-breastfeeding conflict is more acute for women of lower socioeconomic status, such as less-skilled workers and migrant workers, as the employment of these women is concentrated in informal/private sectors, where paid maternity leave benefits are largely unavailable.

DATA

The data used for the present analysis are obtained from the third wave of SCWSS, jointly conducted by the All China Women's Federation (ACWF) and the National Statistics Bureau of China in December 2010. The survey covers a total of 83,940 individuals aged between 18 and 60 years, from thirty-one provinces and province-equivalent municipalities and autonomous regions. Multistage stratified random sampling was applied. In the first stage, 1,300 counties and cities were selected according to the size of the population and the level of development. In the second stage, five urban communities (or villages) were selected from each city (or from a county) using a stratified sampling frame. In the third stage, fifteen households were randomly selected from each urban community (or rural village), and one resident in each household was selected for an interview. Thus, the sample is nationally representative.

The survey collected retroactive information on breastfeeding and maternity leave duration for women's most recent birth. Breastfeeding here refers to any breastfeeding, whether exclusive or complemented with other food, because the survey does not provide information on exclusive breastfeeding. The survey also collected information about the year of the most recent birth, delivery method, child's sex and birth order, and the education attainment of the interviewee and her husband.

Our analysis focuses on employed urban women who gave birth most recently at age 24 or older between 1988 and 2008. Rural women are excluded from the sample because most rural women work on family farmland and are therefore not entitled to maternity leave benefits. We also exclude women whose most recent birth occurred at age 23 or younger, or before 1988, because the late birth bonus policy that we use as an instrument in the empirical analysis was introduced after 1988, and the policy was only applied to those who first gave birth at age 24 or older.⁶ We further restrict the sample to women who had their most recent birth before 2009. This approach minimizes the problem of right censoring in the measurement of breastfeeding duration by allowing each mother at least twelve months for postpartum activities.

Constrained by the lack of a valid instrument for employment status, our analysis focuses on women who were employed at the time surrounding their most recent birth. Because the survey did not collect information on mothers’ employment status at the time of their most recent births, we extrapolate a mother's employment status using other information from the survey. Thus, a mother is considered nonemployed at the time of her most recent birth if this birth occurred before starting her first job, or if she reported having withdrawn from the labor force because of childbirth. These non-employed women are excluded from our sample.⁷ Omitting observations with missing information, the sample for analysis consists of 2,244 women.

EMPIRICAL METHODS

We estimate the determinants of Chinese mothers’ breastfeeding decisions with a discrete choice model below: (1) Equation 1(1) is the structural equation for breastfeeding duration, BF. In this equation, i is the index of mother i, the Greek letters are unknown parameters to be estimated, and u is the random disturbance term. BF is a binary indicator for breastfeeding duration that is equal to 1 if the mother breastfed her baby for at least a given number of months, and 0 otherwise. We consider five BF indicators that are defined based on the thresholds of one, four, six, nine, and twelve months. PMLD denotes paid maternity leave duration, which is measured by the number of days the mother spent on paid leave during the period surrounding her most recent birth. X is a vector of covariate variables that include the mother's age at the most recent birth and its squared term, education attainment of the mother and the father, the child's sex and birth order, a dummy variable that represents whether the mother had a cesarean delivery, and dummy variables for birth-year and regional effects.⁸ The explanatory variables also include the share of public-sector employment in a province for the year of the mother's most recent birth.⁹ This variable is introduced to take into account that women in a locality with a larger public sector may have a higher breastfeeding rate, as they are more likely to have maternity leave benefits and support for breastfeeding from their employers. The definitions and descriptive statistics of the variables included in the regressions are presented in Table A1 in the Appendix.

One major concern about Equation 1(1) is that the variable of paid maternity leave duration (PMLD) may be endogenous because of the unobserved characteristics that simultaneously affect a mother's breastfeeding decision and her maternity leave entitlement. As we show in what follows, maternity leave entitlement in post-reform China is linked to socioeconomic status. However, women of higher socioeconomic status may have a lower preference for breastfeeding. Studies have found that in developing countries where breastfeeding is often perceived as old-fashioned, and bottle-feeding is viewed as a sign of being modern and westernized, there is a negative relationship between breastfeeding initiation rates and duration and socioeconomic status (Rogers, Emmett, and Golding 1997). In addition to the influence of social perceptions, women with higher socioeconomic status can better afford to substitute infant formula for breast milk. These unobserved characteristics would create a downward bias when estimating the effect of paid maternity leave on breastfeeding duration.¹⁰ To control for this endogenous relationship, we use the late-birth bonus days of paid maternity leave in each province plus the common entitlement of ninety days (see Table 1) as an instrument for PMLD.¹¹ The provincial paid leave entitlement (PPLE) is correlated with PMLD, but uncorrelated with a mother's breastfeeding decision when we control for PMLD and other relevant determinants.

The relationship between PMLD and PPLE can be specified as follows: (2) In Equation 2(2) , δ and γ are unknown parameters, and v is the random disturbance term. Equation 2(2) is the first-stage estimation for PMLD, which is endogenous if the random disturbances u and v are correlated. If u and v are independent, there is no endogeneity problem (Wooldridge 2002). The relationship between u and v can be specified as follows: (3) Here, u and v are assumed to be normally distributed with , and hence, the error term of Equation 3(3) , ϵ, is also normally distributed with , where ρ is the correlation coefficient of u and v. In Equation 3(3) , θ is the covariance of u and v, divided by the variance of v. The three equations specified above are estimated as an IV-probit model. The estimate of θ provides a test of the null hypothesis that PMLD is exogenous. For the purposes of comparison, Equation 1(1) is also estimated by probit methods.

To check the robustness of the estimates, we also estimate Tobit and IV-Tobit models in which the dependent variable is the length of breastfeeding (measured in months) and the explanatory variables are the same as those in the probit and IV-probit models.

RESULTS

Descriptive statistics

Table 2 presents summary statistics for paid maternity leave for all mothers in our sample, and by whether a mother has a college education. For the full sample period from 1988 to 2008, 63.3 percent of employed mothers took paid maternity leave at the time surrounding the most recent birth. The leave-taking rate is much higher for mothers with a college education than for those without (89.4 versus 49.5 percent). To cast some light on the changes following the public-sector restructuring in the late 1990s, we divide the sample into two subperiods: 1988–97 and 1998–2008. We note that the proportion of mothers who took paid leave decreased by 6.7 percentage points, from 66.3 percent in 1988–97 to 59.6 percent during 1988–2008. Apparently, much of the decline occurred among mothers without a college education, as the proportion of college-educated mothers who took paid leave only fell by 3.6 percentage points, from 91.6 to 88 percent, while the leave-taking rate among non–college-educated mothers decreased by 21.7 percentage points, from 57.7 to 36 percent. For all leave-taking mothers, the paid leave duration increased by eight days, from 115 days in 1988–98 to 123 days in 1998–2008, and the magnitude of change is more or less the same for the two types of mothers. Despite the expansion of paid leave duration, the average length of paid leave for all employed mothers fell from seventy-three days in 1988–97 to seventy days in 1998–2008 as a result of the decline of leave-taking rates, with the contraction being mainly experienced by mothers without a college education. Specifically, the average length of paid maternity leave for all college-educated mothers decreased slightly from 108 days to 107 days; in contrast, the average length for all non–college-educated mothers fell by twenty-three days, from sixty-two days to thirty-nine days. Consequently, the paid leave duration gap between college-educated and non–college-educated mothers increased from forty-six days in 1988–97 to sixty-eight days in 1998–2008.

Table 2 Paid maternity leave in urban China

	All mothers	Mothers with a college education	Mothers without a college education
Proportion of employed mothers who took paid maternity leave (%)
1988–2008	63.3	89.4	49.5
1988–1997	66.3	91.6	57.7
1998–2008	59.6	88.0	36.0
Duration of paid maternity leave conditional on mothers who took paid leave (day)
1988–2008	119	123	115
1988–1997	115	120	113
1998–2008	123	125	119
Duration of paid maternity leave for employed mothers (day)
1988–2008	72	107	53
1988–1997	73	108	62
1998–2008	70	107	39
Feature of earnings replacement for paid leave (%) (1988–2008)
Proportion receiving essentially the same pay as before	53.7	64.2	44.1
Proportion receiving basic wages	43.0	34.0	51.3
Proportion receiving income allowance	3.3	1.8	4.6
Observations	2,244	775	1,469

Source: All tables are based on data from the third wave of the Survey on Chinese Women's Social Status, 2010.

Why did the proportion of less-educated women without access to paid maternity leave increase in the post public sector restructuring period? This occurred in part because they were more likely to be laid off during the public-sector restructuring, and most laid-off workers were reemployed in the emerging informal or private sector (Du and Dong 2009). In the post-reform period, the informal and private sectors are also the main source of employment for unskilled rural migrants. While it is unfortunate that the survey does not provide information about a mother's occupation and employer at the time of her most recent birth, the sectoral and occupational distributions of the sample at the time the survey was conducted provide valuable insights into the main beneficiaries of paid maternity leave in the post-reform period. In the sample under analysis, 87 percent of women who took paid leave during 1998–2008 were employed in the public sector at the time the survey was conducted, and the remaining 13 percent were employed in the private sector. Moreover, approximately 97 percent of permanent women staff members in government agencies and non-profit public organizations and 85 percent of women employees in SOEs reported taking paid maternity leave after their most recent birth; in contrast, only 37.7 percent of women employed in domestic private firms and 75 percent in foreign firms reported enjoying such benefits. Clearly, the lack of effective means of enforcing maternity leave regulations in the informal/domestic private sectors is a main cause of the declining coverage of paid maternity leave among less-educated working mothers in the post-reform period.

The last part of Table 2 presents statistics on earnings replacement for maternity leave. Of all leave-taking mothers, 44.9 percent reported receiving essentially the same pay as they earned before; 35.9 percent received basic wages; and the remaining 2.7 percent received income allowance. Given that bonuses constitute an appreciable component of total earnings for Chinese workers in the post-reform period, earnings replacement appears more generous for college-educated leave takers than for the non-college educated: 60.6 percent of the former reported receiving the same pay they earned before, while only 33.3 percent of the latter enjoyed this level of earnings replacement. Thus, less educated women were disadvantaged not only in access to paid maternity leave, but also in the level of financial compensation.

Table 3 presents summary statistics for the breastfeeding rates and durations for all mothers in the sample and by education attainment. The statistics in this table indicate that in terms of breastfeeding, loosely defined, urban Chinese mothers on average have high breastfeeding initiation rates and relatively long breastfeeding durations.¹² Among all mothers in the sample, 93 percent breastfed for at least one month, and the rate is similar between leave-taking and non-leave-taking mothers (92.3 versus 94.1 percent). The breastfeeding initiation rate among college-educated mothers is 93.9 percent, slightly higher than the rate of 92.4 percent for non–college-educated mothers. The results reveal no appreciable differences in breastfeeding initiation rates among mothers differentiated by educational attainment and leave taking. The average length of breastfeeding was 10.1 months for all mothers in the sample, 9.4 months for college-educated mothers, and 10.5 months for non–college-educated mothers. Regardless of educational attainment, the duration of breastfeeding is slightly longer for non-leave takers than for leave takers. The seemingly similar breastfeeding patterns between paid leave takers and non-takers suggest that some characteristics of paid leave takers may be inversely related to their inclination to initiate and sustain breastfeeding, offsetting the intrinsic positive effect of paid leave on breastfeeding. We note from Table 2 that the proportion of mothers who breastfed their babies decreased as the length of breastfeeding increased. For all mothers, 86.8 percent breastfed for at least four months, 83.6 percent for at least six months, 65 percent for at least nine months, and 45.4 percent for at least twelve months. A similar pattern is observed for both college- and non–college-educated mothers, although the difference between the two types of mothers increases as the breastfeeding duration lengthens. Compared with women in developed countries, Chinese women have higher rates and longer durations of breastfeeding. For instance, only 45.4 percent of California mothers breastfed for at least six months, and 34.4 percent breastfed for at least nine months (Huang and Yang 2015). As is well recognized, women's breastfeeding decisions reflect the interplay of multiple factors (Dennis 2002). Hence, discerning the impact of paid leave duration requires multivariate regression analysis.

Table 3 Breastfeeding rates and duration in urban China

	All mothers	Mothers with a college education	Mothers without a college education
Breastfeeding initiation rate^a (%)
All mothers	93.0	93.9	92.4
Employed mothers who did not take paid leave	94.1	96.3	93.8
Employed mothers who took paid leave	92.3	93.7	91.1
Duration of breastfeeding (month)
All mothers	10.1	9.4	10.5
Employed mothers who did not take paid leave	11.1	10.1	11.2
Employed mothers who took paid leave	9.6	9.3	9.8
Breastfeeding duration of all mothers (%)
Breastfeeding ≥ 1 month	93.0	93.9	92.4
Breastfeeding ≥ 4 months	86.8	85.2	87.7
Breastfeeding ≥ 6 months	83.6	81.4	84.8
Breastfeeding ≥ 9 months	65.0	60.0	67.6
Breastfeeding ≥ 12 months	45.4	40.9	47.8
Observations	2,244	775	1,469

Note: ^aDefined as proportion of mothers who breastfed their baby for at least one day.

Regression results

Table 4 presents the estimates of the determinants of breastfeeding for at least six months for all mothers, with the probit and IV-probit estimates presented in the first three columns and the Tobit and IV-Tobit estimates in the last three columns. The probit estimates are the focus of our analysis. For the sake of streamlining presentation, we present the probit regression estimates of breastfeeding determination for only one threshold. For the regressions based on other thresholds, we report only the marginal effects of paid maternity leave duration (see Table 5).

Table 4 Estimates of the determination of breastfeeding for at least six months

	Probit	IV-probit		Tobit	IV-Tobit
	Breastfeeding ≥ 6 months	Breastfeeding ≥ 6 months	Paid leave duration	Duration of breastfeeding	Duration of breastfeeding	Duration of paid maternity leave
	(Marginal effect)	(Marginal effect)	(Coefficient)	(Marginal effect)	(Marginal effect)	(Coefficient)
Late-birth leave			0.157**			0.157**
Bonus (PPLE)			(0.0631)			(0.0632)
Paid leave duration	0.000134	0.00391***		0.00223	0.196**
	(0.000114)	(0.000577)		(0.00177)	(0.0853)
Mother's birth age	−0.0267	−0.0322	4.671	−0.294	−1.115	4.671
	(0.0268)	(0.0261)	(5.166)	(0.435)	(1.151)	(5.166)
Mother's birth age^2	0.000335	0.000442	−0.0702	0.00371	0.0160	−0.0702
	(0.000442)	(0.000429)	(0.0854)	(0.00719)	(0.0189)	(0.0854)
Mother's education
High school	−0.0282	−0.131***	29.75***	−0.488	−6.263**	29.75***
	(0.0222)	(0.0238)	(4.156)	(0.354)	(2.686)	(4.156)
College	−0.0483*	−0.228***	51.63***	−0.785*	−10.81**	51.63***
	(0.0262)	(0.0316)	(4.883)	(0.421)	(4.526)	(4.883)
Father's education
High school	−0.0279	−0.0744***	15.01***	−1.145***	−4.008***	15.01***
	(0.0220)	(0.0210)	(4.138)	(0.350)	(1.534)	(4.138)
College	−0.00144	−0.104***	27.01***	−0.679*	−5.862**	27.01***
	(0.0252)	(0.0286)	(4.731)	(0.401)	(2.488)	(4.731)
Child's sex	−0.0326**	0.00213	−5.691*	−0.241	0.898	−5.691*
(boy = 1)	(0.0158)	(0.0189)	(2.987)	(0.252)	(0.807)	(2.987)
Non-cesarean	0.0356**	0.0922***	−18.35***	0.718**	4.285**	−18.35***
delivery	(0.0180)	(0.0178)	(3.564)	(0.302)	(1.739)	(3.564)
Child's birth order	0.0427**	0.110***	−21.67***	1.433***	5.686***	−21.67***
	(0.0184)	(0.0171)	(3.288)	(0.279)	(1.994)	(3.288)
Proportion of employment in public sector	0.345***	−0.119	72.64***	14.38***	−2.287	72.64***
	(0.118)	(0.177)	(23.87)	(1.965)	(8.827)	(23.87)
Birth year dummy	Y	Y	Y	Y	Y	Y
Region dummy	Y	Y	Y	Y	Y	Y
θ/(1-ρ^2)		−1.357***
		(0.474)
α					−0.195**
					(0.0853)
Wald test of exogeneity		chi2(1) = 8.21			chi2(1) = 5.21
		P value = 0.0042			P value = 0.023
Log likelihood	−959.772	−13654.667		−497.154	−19518.432
Observations	2,244	2,244	2,244	2,244	2,244	2,244

Notes: Heteroscedasticity-robust standard errors are presented in parentheses. ***, ** and * denote statistical significance, at the 1, 5, and 10 percent levels, respectively. θ/(1-ρ²) and α are a measure of the correlation between the error terms of the first-stage and the second-stage equations.

Table 5 Effects of paid maternity leave duration on the probability of breastfeeding for at least 1, 4, 6, 9, and 12 months

	Marginal effects IV probit^1	Effects of increasing paid leave by 30 days	Effects of decreasing paid leave by 23 days for mothers without a college education
Breastfeeding ≥ 1 month	0.0033***	0.099***	−0.076***
	(0.0013)
Breastfeeding ≥ 4 months	0.0038***	0.114***	−0.087***
	(0.0007)
Breastfeeding ≥ 6 months	0.0039***	0.117***	−0.090***
	(0.0006)
Breastfeeding ≥ 9 months	0.0043***	0.129***	−0.099***
	(0.0002)
Breastfeeding ≥ 12 months	0.0043***
	(0.0001)	0.129***	−0.099***
Observations	2,244

Notes: Marginal effects of IV-probit equations are presented with heteroscedasticity-robust standard errors in parentheses. Each equation controls for the same covariates as the equations presented in Table 4. ***, ** and * denote statistical significance at the 1, 5, and 10 percent levels, respectively.

With respect to the IV-probit model, we first examine the estimates of Equation 2(2) , the reduced-form equation of PMLD. The estimates show that the variable PMLD is positively correlated with the instrumental variable PPLE, and this correlation is significant at the 5 percent level. With respect to the covariates of the reduced-form equation, the estimates show that, other things being equal, the length of paid leave is positively correlated with both a woman's education level and her husband's education level. This result perhaps captures the influence of assortative marriage norms. The estimates also show that having a non-cesarean delivery or a child in higher birth order reduces paid leave duration. As we would expect, provinces with higher shares of public-sector employment have longer paid leave durations. As a measure of the correlation of the error terms of Equations 1(1) and 2(2) , the estimates of the term θ/(1-ρ²) presented at the bottom of Table 4 indicate that the two error terms are negatively correlated, and the relationship is significant at the 1 percent level. The negative correlation implies that there are unobserved characteristics that are positively associated with a mother's paid leave entitlement and negatively associated with her inclination to sustain breastfeeding. The failure to control this unobserved heterogeneity will generate a downward bias in the estimation of paid leave effects. In line with the estimates of θ/(1-ρ²), the Wald test rejects the null hypothesis that PMLD is exogenous at the 1 percent significance level. Thus, the probit estimates are biased and inconsistent, while the IV-probit estimates are consistent. Evidently, the probit estimates of paid leave duration for Equation 1(1) are much smaller than the IV-probit estimates, and the two types of estimates differ noticeably for most of the covariates of Equation 1(1) .

We now closely examine the IV-probit estimates of Equation 1(1) . As expected, PMLD has a positive effect on the probability that a mother breastfed for at least six months, and the effect is significant at the 1 percent level. Numerically, other things being equal, a ten-day extension of paid leave increases the probability of breastfeeding for at least six months by 3.9 percentage points. Turning to the covariates, we note that other things being equal, breastfeeding rates are negatively correlated not only with a woman's education but also with her husband's education. Compared with the probability for women with a middle school education or lower, the probability of breastfeeding for at least six months is 13.1 percentage points lower for women with a high school education and 22.8 percentage points lower for women with a college education. Similarly, compared with the results for women who are married to husbands with a middle school education or lower, the probability of breastfeeding for at least six months is 7.4 percentage points lower for those whose husbands have a high school education and 10.4 percentage points lower for those whose husbands have a college education. The negative relationship between breastfeeding rates and women and their husband's educations is congruent with the finding that breastfeeding initiation rates and duration are inversely correlated with women's socioeconomic status from the studies of other developing countries mentioned previously. Moreover, the estimates show higher rates of breastfeeding in women who had a non-cesarean delivery or a child of higher birth order. These results may be attributed to the fact that cesarean delivery hinders a mother's ability to produce breast milk and mothers paid more attention to the second born than the first one. Finally, the probit estimate for the share of public-sector employment in a province has a positive sign and is statistically significant. However, it becomes insignificant in the IV-probit estimation in which its correlation with PMLD is controlled for in the first-stage estimation.

As a robustness check, we present the Tobit and IV-Tobit estimates in the last three columns of Table 4. The reduced-form equation of PMLD in IV-Tobit estimation is the same as in IV-probit estimation. The estimate of α parameter again reveals a significant negative correlation between the error terms of the first-stage and second-stage equations. Moreover, the Wald test rejects the null hypothesis that PMLD is exogenous to the length of breastfeeding at the 5 percent significant level. The estimate of PMLD is significant at the 5 percent level, indicating that if PMLD increases by one day, breastfeeding duration will increase by 0.196 month. The IV-Tobit estimates of the covariates are substantively similar to the IV-probit estimates. Thus, our regression results are robust.

Table 5 presents the IV-probit estimates of the marginal effects of paid leave durations on the probability of breastfeeding for at least one, four, six, nine, and twelve months. For all five thresholds, the IV-probit estimates have a positive sign and are significant at the 1 percent level. The effect of paid leave duration on breastfeeding rates increases with the breastfeeding duration threshold; numerically, a thirty-day extension of paid leave will increase the probability of breastfeeding for at least one, four, six, nine, and twelve months by 9.9, 11.4, 11.7, 12.9, and 12.9 percentage points, respectively. These estimates support the claim that by safeguarding women's employment and income security during childbirth and breastfeeding, paid maternity leave enhances the ability of employed women to sustain breastfeeding. Using the estimates, we further explore the consequences of the declining paid leave entitlement for mothers without a college education. As is indicated in Table 2, from 1988–97 to 1998–2008, the average length of paid leave for non–college-educated mothers decreased by twenty-three days. This change reduced the probability that mothers without a college education breastfed for at least one, four, six, nine, and twelve months by 7.6, 8.7, 9, 9.9, and 9.9 percentage points, respectively. Since the average length of paid leave for college-educated mothers remained more or less unchanged, the decrease in non–college-educated mothers’ breastfeeding rates also mean that holding other factors constant, the breastfeeding gap between mothers with and without a college education has increased in the post public-sector restructuring period. Thus, paid leave disparities have important implications for intergenerational inequality.

Notably, our estimates of paid maternity leave effects on breastfeeding duration are strikingly compatible with the estimates obtained for California by Huang and Yang (2015). Under the paid family leave regulation in California, eligible women workers are entitled to a maximum of six weeks of partially paid leave to bond with a new child. Huang and Yang (2015) estimated that the adoption of this regulation has increased the probability of breastfeeding for at least six months by 21 percentage points and increased the probability of breastfeeding for at least nine months by 19 percentage points. According to the IV-probit estimates presented in Table 5, a six-week extension of paid leave duration increases the probability that an urban Chinese mother breastfeeds for at least six and nine months by 16.4 and 18.1 percentage points, respectively.

CONCLUSIONS

Using data from a nationally representative survey, this paper estimates the effect of paid maternity leave on breastfeeding duration in urban China during the period from 1988 to 2008. The analysis shows that following the public-sector restructuring, the proportion of women who took paid maternity leave after their most recent birth declined, and much of this decline was endured by women without a college education, who were primarily employed in the informal or private sectors. Despite the rising disparity in maternity leave entitlements, the breastfeeding patterns of women without a college education were fairly similar to the patterns of their more educated sisters whose access to paid leave has not been adversely affected by market reforms. The analysis disentangled the complex relations between paid maternity leave entitlements and breastfeeding decisions by exploring variations in the late-birth bonus policy across provinces. The analysis shows that paid maternity leave has a positive effect on breastfeeding duration, and the decline of paid leave entitlement among non–college-educated mothers decreased their inclination to sustain breastfeeding and widened the breastfeeding gap between mothers with and without a college education.

The findings reported in the present paper have important policy implications. They underscore the importance of social protections for women's reproductive role, especially for women with low socioeconomic status in China's new market economy. Expanding coverage of paid maternity leave in the informal/private sectors should thus be a policy priority. Universal paid leave entitlements would not only reduce income disparities among women in different socioeconomic groups but also minimize intergenerational inequality. By making maternal employment more compatible with breastfeeding, the expansion and improvement of paid maternity leave programs are also essential to boost fertility rates and slow the population aging process. In late 2013, the Chinese government replaced the long-standing one-child policy with a new policy that allows a couple to have two children if one of the parents is an only child. However, by October 2015, only 16.8 percent of eligible couples in the cities had submitted applications to have a second child. In 2016, a new policy that encourages all couples to have two children was adopted. Without a substantial increase in social support and protection for working mothers, Chinese women's willingness to have more children will remain low.

Acknowledgments

The work was carried out with the aid of a grant from the International Development Research Center of Canada (Project no. 107579) and a grant from the National Natural Science Foundation of China (Grant no. 71303189).

Additional information

Notes on contributors

Nan Jia

Nan Jia is Professor in the Department of Economics at the Southwestern University of Finance and Economics, China. Her research focuses on China's economic development, with an emphasis on labor and gender issues.

Xiao-yuan Dong

Xiao-yuan Dong is Professor in the Department of Economics at the University of Winnipeg, Canada, and an adjunct Professor of the National School of Development, Peking University, China. She has served on the boards of directors of the International Association of Feminist Economics and Chinese Economists Society. She is an Associate Editor of Feminist Economics and has served on the editorial boards of China Economic Review, Journal of Comparative Economics, and Journal of Socio-Economics. Her research focuses on China's economic development, with an emphasis on labor and gender issues.

Yue-ping Song

Yue-ping Song is Associate Professor at the Research Center of Population Development at Renmin University, China. Her research interests focus on migration, gender inequality, and child health in China. Her work has been published in Research in Labor Economics and Journal of Development Studies.

Notes

1 Several studies have empirically explored the link between paid maternity leave and breastfeeding without controlling for the potential endogeneity between the two variables. For instance, from a sample of 770 full-time working mothers in California, Sylvia Guendelman et al. (2009) found a positive association of paid maternity leave length and breastfeeding initiation rates and duration. Using data from a nationally representative survey, Chinelo Ogbuanu et al. (2011) found that when other factors were controlled for, the length of paid maternity leave had no effect on breastfeeding initiation rates and breastfeeding duration, whereas the delay of the return to paid work lengthened the duration of breastfeeding among mothers in the United States.

2 A selective list of studies on this topic includes Liu, Dong, and Zheng (2010), Maurer-Fazio et al. (2011), Du and Dong (2013), and Jia and Dong (2013).

3 State Council, Special Regulations for Labor Protection for Women Workers. http://www.gov.cn/zwgk/2012-05/07/content_2131567.htm.

4 In the post-reform period, labor compensation for most Chinese workers consisted of basic wages and bonuses.

5 The result on breastfeeding rooms is from a survey conducted by the Chinese Enterprise Association (CEA) in 2005 (Liu, Zhang, and Li 2009). The statistic on nurseries and kindergartens is from the Chinese Enterprise Social Responsibility Survey conducted by The China Center for Economic Research of Peking University and the World Bank (Du and Dong 2013).

6 Since the one-child policy was formally introduced nationwide in 1982, the most recent birth is also the first birth for most women in our sample.

7 Ideally, self-employed women who are not entitled to maternity leave benefits should also be excluded from our sample. Unfortunately, the survey does not provide information on a mother's occupation at the time of her most recent birth.

8 The survey does not provide information on where the most recent birth occurred. To control for unobserved regional effects, we divide the thirty-one provinces into three regions – eastern, central, and western – based on where the mother lived at the time of interview.

9 This variable is defined as the ratio of the number of staff and workers in SOEs and urban collectives to the total number of staff and workers in a province in the year of the mother's most recent birth. The data source is the China labor statistical yearbook.

10 We are unable to control for income effects because of the lack of information on family incomes at the time of the most recent birth.

11 The late-birth bonus policy was enacted in different months. The monthly differences are aggregated into yearly differences when the PPLE is derived. Specifically, if the policy was enacted in the first six months of a given year, the corresponding year is chosen as the year in which the policy was implemented. If the policy was enacted in the second half of a year, the next year is used as the year of policy implementation.

12 According to the Fourth National Health Services Survey conducted by the Center for Health Statistics and Information in China's Ministry of Health in 2008, only 27.6 percent of Chinese mothers exclusively breastfed their babies through the first six months, and the six-month rate of exclusive breastfeeding was 15.8 percent for urban mothers and 30.3 percent for rural mothers.

Appendix

Table A1 Description and descriptive statistics of the variables used in the analysis

Variable	Definition	Mean
Breastfeeding initiation	Dummy = 1 if the mother breastfed for a month or longer; 0 otherwise	0.930
Breastfeeding duration	Length of time, in months	10.879
		(5.185)
Paid maternity leave duration	Length of time, in days	71.739
		(79.349)
Provincial paid leave entitlement	Provincial paid maternity leave bonus for late birth plus 90, in days	114.870
		(26.834)
Mother's birth age	Mother's age when gave the most recent birth	27.635
		(3.417)
Mother's age	Mother's age when interview	42.359
		(8.179)
Mother's education
Middle school or lower	Dummy = 1 if the highest level of education is middle school or lower; 0 otherwise	0.328
High school	Dummy = 1 if the highest level of education is high school; 0 otherwise	0.326
College or higher	Dummy = 1 if the highest level of education is college or higher	0.345
Father's education
Middle school or lower	Dummy = 1 if the highest level of education is middle school or lower; 0 otherwise	0.294
High school	Dummy = 1 if the highest level of education is high school; 0 otherwise	0.321
College or higher	Dummy = 1 if the highest level of education is college or higher	0.385
Non-cesarean Delivery	Dummy = 1 if the birth was a non-cesarean delivery; 0 otherwise	0.731
Child is a boy	Dummy = 1 if the child was a boy; 0 otherwise	0.580
Child's birth order	Equal to 1 if the child was first born, and equal to 2 if the child was second born	1.253
		(0.532)
Share of public-sector employment in a province		0.727
		(0.230)
Observations		2,244