First Draft
Does a Mother’s Early Return to Work After Childbirth Improve Her Future Employment Status?: A Quasi‐experiment Using Japanese Data1 28 December 2015
Wataru KUREISHI, Colin MCKENZIE, Kei SAKATA and Midori WAKABAYASHI
1
The authors would like to gratefully acknowledge the financial assistance provided by the Japan Society for the
Promotion of Science (JSPS) Grant in Aid for Scientific Research (B) No. 23330094 for a project “Life Events and Economic Behaviour: A Perspective of Family Inter-dependence” (Project Leader: Midori Wakabayashi). Data used in this paper comes from the Japanese Ministry of Health, Labour and Welfare’s Longitudinal Survey of Newborns in the 21st Century (21 seiki shussho ji judan chosa) and the Live Birth Form of Vital Statistics (Jinko dotai chosa shusseihyo). The fourth author wishes to acknowledge the financial assistance provided the Tohoku Kaihatsu Memorial Foundation and the Nomura Foundation.
1
First Draft Abstract The purpose of this paper is to examine empirically whether or not a mother’s early return to work after maternity leave and parental leave leads to a higher working status. This paper attempts to find out if a mother returns to work within 1 year after childbirth, then she is more likely to stay employed and is more likely to work as a full-time worker than to work as a part-time worker. We estimate 2SLS and bivariate probit models using July births as an instrument. Our IV approach is unique in that we shed the light on the relationship between the timing of birth (birth month) and the timing of the mother’s return to work. Our empirical evidence finds that an early return to work has a positive causal effect on the likelihood of a mother being in fulltime employment in the long-term. However, the positive causal effect of an early return to work on being employed including part-time, self-employed or family worker, and pieceworker at home is not observed.
2
First Draft
1. Introduction Japanese society is now rapidly ageing due to a long-term decline in the birth rate. This in turn means a rapid decrease in the labor force. In order to deal with the decline in fertility rate and the rapid decrease in the labor force, Prime Minister Shinzo Abe has proposed various policy measures. One of his most contentious policy measures is an extension of childcare leave from 1.5 years to 3 years2. His intention is to encourage mothers to stay in the labor force after childbirth and childcare and to reduce the number of children on the waiting list for admission into child care facilities. As he praised himself in his speech, it is good thing that we give babies hugs and cuddles as we like itself. As the proposed policy intends, the longer childcare leave may encourage women to return to work as their full-time position is secured. However, some people argue that the policy may in fact have the opposite effects to what is intended. They criticize the policy by arguing that longer childcare leave deprives mothers of their job skills and career accumulation as the longer they are out of the labor force, the more likely their job skills depreciate. Therefore, we focus on this argument, that is, the earlier mothers return to work after childbirth, the less likely that they experience a depreciation of their job skills, and the more likely it is that they are in a full-time position. Previous literature has so far been made in the context of identifying policy intervention of childcare and parental leave in the settings of difference-in-differences and regression discontinuity design. In Japan, Yamaguchi (2015), who reviews research situation on childcare and parental leave policy and female employment in Japan, says that very few papers from Japan in this literature seriously address the identification issues.3 In Germany, where several reforms in maternity and childcare leaves have been gone through since 1970s, Schonberg and Ludsteck (2007) compared women who gave birth just before and just after reforms. They say that
2
Speech on Growth Strategy by Prime Minister Shinzo Abe at the Japan National Press Club, Friday, April 19,
2013 3
The few exceptions are Asai (2015) on the replacement rate of childcare leave and Nagase (2014) on reform on
short-hour option for employee with children under three.
3
First Draft each expansion in maternity leave delays mothers’ return to work, but had little impact on mothers’ labor force participation in the long run. Geyer et al (2014) focused on the 2007 reform of parental leave benefit and find modest positive effects on the labor supply of mothers in the second year after their child was born. In Austria, Lalive et al (2014) and Lalive and Zweimüller (2009)) showed that extended parental leave significantly reduces return to work but employment and earnings decrease in the short run but not in the long run. In US, Baum (2003) shows that 12 week unpaid maternity leave legislation has small and statistically insignificant effects on employment and wages. In Canada Baker and Milligan (2008) find that the introduction and expansion of statutory maternity leaves increase the likelihood of mothers’ returning to pre-birth employer and job continuity. Alternative approach is the strategy using quarter of birth as instrumental variables in line with Angrist and Kruger (1991) and Gelbach (2002). Most recently, Bauernschuster and Schlotter (2015), using day-of-birth cutoff dates as instruments for kindergarten attendance, find significantly positive causal effects on mothers’ employment probability and on their weekly working hours in Germany. Another closely related study is Berlinsli et al (2011), who found no effect of preschool attendance on maternal labor outcome using enrollmentage rule. By using data from the Longitudinal Survey of Newborns in the 21st Century, this paper examines whether a mother returns to work within 1 year after childbirth has any impact on being employed and being in full-time employment. We estimate bivariate probit models with July births as an instrument. Method is a quasiexperiment using birth month of newborn babies as an instrumental variable. Our approach is notable in that we direct attention to the relationship between the birth month of a new born baby, the deadline for admission in a licensed child care facility, and mother’s early return to work. The data we use are a panel data of Japanese newborns and parents, which gives us newborns‘ birth month. As we will explain later in detail, the birth month of a baby is crucial in determining the timing of a mother’s return to work due to the application deadline for childcare facilities. Our empirical evidence finds that an early return to work after child-birth has a positive causal effect on the likelihood of mother’s full-time employment in both the short-term and the long-term. However, the positive causal effect of an early return to work on being employed is shown only in short-term.
4
First Draft The rest of this paper is organized as follows. Section 2 provides details of the key institutional features of the childcare leave system in Japan. The estimation method and the data used in this paper are discussed in sections 3 and 4, respectively. Section 5 discusses the use of the baby’s birth month as an instrument variable, while section 6 presents the estimation results. The credibility of the exclusion analysis used in this paper is discussed in section 7, and section 8 concludes the paper.
2. Parental Leave and Childcare Facilities in Japan In Japan, an expectant or new mother is eligible for two types of legal leave: maternity leave and parental leave. For maternity leave, any working woman is permitted to take 6 weeks of leave before delivery (14 weeks in the case of multiple pregnancy) and 8 weeks of leave after delivery. The leave before delivery is available upon request, while the leave after delivery is mandatory. Public health insurance pays a maternity allowance to a working woman who takes maternity leave. In general, two thirds of the standard daily income are paid per day. In 2001, fathers and mothers who are workers, excluding daily and fixed-term employment workers, could take childcare leave until their child reaches the age of one upon request.4 Unemployment insurance paid 40% of the standard income as a childcare leave benefit.5 In addition, for the periods of the maternity and parental leaves, workers were exempted from paying both out-of-pocket and employer’s contributions of social security. An increasing number of women take childcare leave while it is extremely rare for men to take parental leave in Japan. The Basic Survey of Women Workers' Employment Management shows that in 2002 64.0% of mothers took childcare leave while only 0.33% of fathers did (Ministry of Health, Labour and Welfare, 2002). 4
From 2005, fixed-term employment workers who (a) have been continuously employed one year or more by the
same employer and (b) are expected to continue their employment after their child’s first birthday have been able to take childcare leave. 5
At 2007, the percentage went up to 50%.
5
First Draft These maternity and parental leaves enable many full-time working mothers to return to work after their children reach the age of one. In Japan, traditionally, the use of child care service was not an option for parents and child rearing was entirely on the sholder of parents and their relatives. However, due to increases of working mothers, the demand for child care facilities has increased over the years. Thus, child care facilities in Japan are set up under the provision of the Child Welfare Law. In the period of high economic growth of 1960s, child care facilities became available in response to the diversification of childcare needs, which stem from increases in the number of working mother. According to the Report on Social Welfare Administration and Services by the Ministry of Health, Labour and Welfare, in 2002 for 0-year-old children only 6% are left in child care facilities in Figure 1. As the age of the child rises, the ratio of children nursed in child care facilities increases 21% of 1-and-2year-old and 36% of 3-year-old children are nursed in child care facilities. The ratio for 4-year-or-older children increases slightly, which may be because children of 4 years or more start going to kindergarten. For the most recent year, 2014, the ratio for 0-year-old children has increased to 15%. The main child care facilities are licensed child care facilities (Ninka Hoikusho) subsidized by local governments. Due to subsidies, licensed child care facilities are much cheaper than non-licensed facilities. Licensed child care facilities are regulated by the Child Welfare Act (Jido Fukushi Ho). In order for a child care facility to be licensed by a local government, the child care facility must fulfill certain minimum standards set by the Child Welfare Act relating to as the size of building, the size of playground, the number of nursery teachers, the content of the nursery program, and nursery hours. The hours of child care services are 8 hour per day. The operating expenses of these licensed child care facilities are subsidized by the local government, and the childcare fees paid by parents are based on the family’s income. According to the 2003 Survey of Regional Child Welfare Services, 29.7% of households with one child paid 20,000 to 30,000 yen for childcare, while 22.8% paid less than 10,000 yen. For households with two children, 16.4% paid 20,000 to 30,000 yen for childcare, while 15.8% paid less than 10,000 yen, and 30,000 yen to 40,000 yen, respectively. 6
First Draft Non-licensed child care facilities, those facilities which are not authorized under the Child Welfare Law, have complementary functions to the licensed facilities in the present childcare system, and provide various kinds of childcare services which are not available from the licensed child care facilities, such as childcare immediately after maternity leave, extended hours of childcare service, nighttime childcare, and accepting children halfway through the education year. However, as non-licensed child care facilities hardly receive any subsidies from the local government, many of them are financially constrained, and it is often said that some provide inferior childcare services to licensed facilities. Due to their lower fees and relatively higher quality, parents with newly born child(ren) prefer to use a licensed facility. However, not all children can be admitted to the licensed child care facilities due to the limited number of facilities. It is important to note that under the Child Welfare Act, licensed child care facilities are provided to those guardians who are unable to look after their child(ren) at home during the daytime for reasons such as work and disabilities. In other words, unless both parents work, their child(ren) are not eligible for admission into a licensed child care facility. There are two important features of licensed childcare facilities that are relevant for our analysis. The first is that licensed childcare facilities only accept new children once a year on 1 April, the beginning of the academic and business year in Japan. The second is that the application deadline for an April admission in licensed child care facilities is in December of the previous year.
3. Estimation Method Our objective is to estimate a causal model which describes whether or not a mother’s early return to work after childbirth leads to a higher working status. It is represented by the following equation: + (1) 7
First Draft where the outcome variable Y is mother’s future employment status as a full-time worker, the treatment variable T is mother’s early return to work after child birth, X are the vector of control variables, and u is the error term. In order to identify the causal inferences, we have to overcome an econometric problem. That is, treatment variable T may be correlated with the error term, which biases the estimates of the regression coefficients. The first concern is that the mother’s future employment status and her early return to work may share some unobservable common causes which induce spurious associations, or confounding bias. On the one hand, employers may place mothers with high abilities into higher positions (full-time positions) due to their abilities and be generous to take longer period of childcare leaves. In this case, future employment would show a spurious downward bias to the coefficient of an early-return-to-work variable despite these two variables being independent. On the other hand, mothers with high abilities may want to return to work earlier due to their opportunity costs, which causes an upward bias. Which bias dominates the other is ambiguous. Thus, we will overcome the endogeneity by a quasi-experiment using baby’s birth month as an instrument for treatment variable T. In addition, we also conduct the following bivariate probit model: 1
0 and
1
, (2)
where vi is independent of X and X includes instruments of children’s birth month we will explain in the following section 5.
8
First Draft
4. Data The data we use is from the Longitudinal Survey of Newborns in the 21st Century (21 seiki shussho ji judan chosa, in Japanese), which is the first longitudinal survey conducted by the Japanese Ministry of Health, Labour, and Welfare.6 The survey is a mail-in longitudinal census survey which has tracked all newborn babies born in Japan in the periods of January 10-17 in 2001, and in the period of July 10-17 in 2001, and has been conducted every year since 2001.7 The first survey which we refer to as the 2001 survey was conducted when the babies were 6 month old: August 1 2001 for the January babies and February 1 2002 for the July babies. The number of questionnaires delivered and responses received for the 2001 survey are 26,620 and 23,421 for the January babies (a response rate 88.0%), and 26,955 and 23,589 for July babies (a response rate of 87.5%), respectively. Our instrumental variable Z is a dichotomous variable which equals one if the baby born in July 2001 and zero otherwise (that is, babies born in January 2001). For the output variable Y, we measure a mother’s employment status using three variables. The first one is a dichotomous variable which equals one if the mother is working as a full-time worker and zero otherwise. The second is also a dichotomous variable which equals one if the mother is working (as a full-time, part-time, self-employed or family worker, or pieceworker at home), and zero otherwise. The third is a continuous variable which measures the mother‘s annual earned income. , which was asked only at 2002, 2004, 2005, 2008, and 2011 surveys. For the treatment variable T, we use a question in 2001 survey, which asks only for respondents who were employed as full-time workers six months after delivery. There are three sub-categories for respondents who answered “Yes” to this question: (1) those who have already taken childcare leave; (2) those who are currently on childcare leave; and (3) those who plan to take childcare leave. Those respondents who chose one of
6
7
Sakata et al (2015) gives a summary of the survey (in Japanese). The newborn babies were sampled from the Live Birth Form of Vital Statistics (Jinko dotai chosa shusseihyo).
9
First Draft these three categories were then asked about the (expected) length of their childcare leave (in months). For respondents who answered “No” to the question about taking childcare leave, there are three sub-categories: (4) their workplace has a childcare leave system but they did not take it up; (5) their workplace does not have a childcare leave system;8 and (6) those who do not know whether or their workplace has a childcare leave system. The question explicitly instructed respondents to exclude 8 week maternity leave and short-term working. Because, as we mentioned earlier, only those who are expected to continue their work after their children’s first birthday are eligible for childcare leave, we suppose that they return to work after their childcare leave. So we construct the treatment variable T representing a mother’s early return to work less than 12 months from childbirth as a dichotomous variable which equals one if the length of mother’s childcare leave is 9 months or less, and zero otherwise.9 Our sample is selected from the original 47,015 (23,421 January babies and 23,589 July babies) as follows. First, we use only respondents whose babies are first-born because mothers with more than one child may have more and better knowledge of child-bearing and child-rearing. They may also manipulate the timing of their child’s birth given knowledge that having a baby in the period from January to March is disadvantageous for child care admission. This reduces the sample to 23,503. Second, we restrict the sample to respondents who were employed as full-time workers both one year before delivery and a half year after delivery. This is because workers who have been continuously employed one year or more by the same employer have been able to take childcare leave and because the questions about taking childcare leave and the length of any childcare leave were asked only to those mothers who respondents who were employed as full-time workers when the 2001 survey was conducted. This reduces the sample to 4,088. Finally, restricting the sample to those who answer other questions providing necessary information reduces the final sample to 3,151.
8
9
As an alternative, we use the length of mother’s childcare leave (in months) as a continuous treatment, but we
do not have …
10
First Draft Descriptive statistics for this sample of above mentioned full-time working mothers using information mainly from second survey, the 2002 survey, are presented in Table 1. At the time of the 2002 survey, when the newborn babies were one and a half years old, 83 percent of mothers worked as full-time workers and 87% worked in other employment statuses such as part-time, selfemployed, and domestic-piecework. Mothers were on average 30.6 years old, which means that for our sample the average age of bearing the first child is 29 years. 23.8% of mothers have university degree or more. The average number of siblings is 0.07 which indicates that some of the babies born in 2001 have younger brothers or sisters. In the previous year (that is, 2001), husbands earned about 4.39 million yen annually (including tax) on average, which is 0.2 million yen higher than one year earlier (2000). In Figure 2 (a), we can observe mothers’ employment status at the time of the 2004 survey when the newborn babies have reached the age of three and a half. If a mother had returned to work earlier after childbirth, that is, the length of childcare leave was 9 months or less, 76.8% of them worked fulltime and 7.5% worked parttime at 2004 survey, while if a mother returned to work 10 or more months childcare leave, 70.2% worked fulltime and 7.7% worked part-time. In addition, only 14.2% of mothers returning to work early were not working three years after their childbirth, while nearly 20.3% of those not returning to work early were not working three years after their childbirth. Figure 2 (b) shows mothers’ employment status a decade after childbirth. Although the ratio of fulltime workers for those who returned to work early is lower than three years after childbirth and that for those who had not return to work early also decrease, there still persists a gap in the ratio of mothers working fulltime between those who returned to work early and those who did not. In addition, because children aged 10 might require less attention, compared to what is observed here years after childbirth the ratio of those who did not work decreases and that of those who work as part-time workers increases, regardless of returning to work early.
11
First Draft
5. IV: Baby’s Birth Month It is important to stress that the timing of a birth is crucial for the admission of the child into a licensed childcare facility. In many municipalities, infants can enter a 0-year–old childcare class of a child care facility after they turn 3 to 6 months. In Japan, the school calendar year including that for licensed childcare facilities starts from April, and the application deadline for April admissions in licensed child care facilities is in December of the previous year. Thus, for babies born between January and March, the chance for the admission in a licensed child care facility is very slim as they miss the deadline December of the previous year. In other words, those babies born between January and March (together with babies born in April or later of the same year) have to wait until April of the following year for the admission in a licensed child care facility. This means that mothers who give birth from January to March have to either find a non-licensed facility, which is more expensive, or a babysitter such as a grandparent when their childcare leave expires. On the other hand, those who give birth after April in the same year can return to their full-time position relatively smoothly after 12 months of maternity leave and childcare leave as they can make it to their first deadline in December of the year for the admission in a licensed child care facility. Since strict a municipality’s implementation of the application deadline for child care facility admissions are keyed to a child’s birth month being before/after April, mothers who give birth after April can return to work earlier after childbirth than those who give birth from January to March. For example,
Taro, who born January, 2001, cannot enter 0-year-old class because the deadline was already passed at December 2000, but enters 1-year-old class in April 2002, making his mother to return to work after more than 1 year after childbirth.
Hanako, who born July 2001, can enter 0-year-old class in April 2002 because the deadline is December 2001, which make her mother to return to work after less than 1 year after childbirth.
12
First Draft If giving birth between January and March does not affect mother’s employment status through any mechanism except mother’s early return to work, the timing of birth is a valid instrumental variable for the effect of early return to work. According to the Comprehensive Survey of Living Conditions, regarding the status of childcare by children’s age, 70.9% of babies aged 0 of mothers with job are taken care of by father and mother, while for children aged 1 to 6 the percentage of licensed child care facilities exceeds 50%. The percentages of grandparents and unlicensed child care facilities are 10% or so, respectively. Children’s birth month Z randomizes not mothers’ return to work but the eligibility of the child care facility admission, so it introduces an element of randomness in the process of child care facility admission. That is, in our quasi-experimental setting, we exploit the random variation in mothers’ early return to work introduced by our quarter-of-birth type instrument (Angrist and Kruger (1991)). This gives us the following first stage equation: . (3) The instrument relevance is presented by the coefficient
on children’s birth month, which indicates the share of
mothers who return to work early according to their children’s birth month. Figure 3 indicates how the distributions of the lengths of childcare leave in months differ between mothers who gave birth in January and those in July. The most important finding is that mothers who gave birth on July are more likely to take childcare leaves 6 to 9 months than mothers who gave birth on January (for 7 months of childcare leave, 9.1% for those who gave birth on July and 2.4% for those on January), while mothers who gave birth on January are more likely to take childcare leaves 10 and 12 months than mothers who gave birth on July (for 10 month 16.5% for July and 22.5% for January, and for 12 months 24.7% for July and 27.6% for January). This implies that mothers who gave birth on July return to work in April (6 or 7 months after maternity
13
First Draft leave), which corresponds to the beginning of the school year. That is, they have better chance to admit their children in licensed child care facilities than those on January. Another interesting pattern is that about 50% of mothers who gave birth on January and about 40% of those on July answered 10 to 12 months childcare leave. There are few fulltime working mothers who took (or expected to take) childcare leave more than 13 months because an eligible working mother can collect childcare leave benefits until her child reaches the age of 1.
6. Credibility of the Exclusion Restriction One possibility that leads to bias in our IV estimator is that the instrumental variable is confounded with socio economics status (SES). For example, Buckles and Hungeman (2013) show that controlling for parental SES reduces the association between the quarter of birth and wages. Their paper argues against Angrist and Kruger’s (1991) proposal of using the quarter of birth as an IV. Buckles and Hungeman show that the mothers of children born early in the year are younger, have less education, and are less likely to be married or white. These factors directly affect wages So we also checked the difference in parents socio economic factors before childbirth, such as fulltime working one year before delivery, weight at birth, mother’s and father’s education, and so on between January and July births. As can be seen if Figure ?, for families where the mother was working fulltime before delivery, there are no statistically significant differences in the socio-economic factors.
14
First Draft
7. Results 1. The Results of Estimations for Mothers’ Fulltime Working in 2004 In Table 2 we show the results of OLS, probit, 2SLS and bivariate probit estimations of outcome variable Y (mother’s fulltime working) and treatment variable T (early return to work after childbirth) in year 2004. In the table we present the coefficients for OLS, 2SLS, and bivariate probit and marginal effects for probit. Columns (a) and (b) are estimated results of OLS and probit models, respectively, where endogeneity is not taken account for. The results suggest that treatment variable T, early return to work after childbirth, is positive and statistically significant. That is, a mother who returns to work early within 9 months after childbirth are 11.2% and 11.4% points more likely to work as fulltime workers in 2004. In columns (c) and (d), we present the results of 2SLS model, where both the outcome variable Y and treatment variable T are assumed to be linear. Our first stage estimation in columns (c) gives us that positive and statistically significant coefficient of the instrumental variable, July. That is, a mother who gave birth on July is 7.7% more likely to return to work within 9 months after childbirth. We could not reject the null hypothesis of Sargan test for over identification. For the other instrumental variables, we have negative and significant coefficients of valid job vacancy rate. That is, mothers who live in the prefecture with one percentage point higher valid job vacancy rate are 21.5 % less likely to return to work earlier within 9 months after childbirth. For control variables, we have positive and significant coefficients of mother’s university degree. That is, mothers who have university degree are 5.5% points less likely to return to work after childbirth. In addition, the coefficient of father’s log of annual income in the previous year 2003 is negative and significant. That is, if fathers earn more in the previous year, then mothers are less likely to return to work earlier. Moreover, we have positive and significant coefficients of smaller firm size and a negative and significant coefficient of firm size being public sector. That is, mothers who work for larger firms are less likely to return to work earlier. Moreover, we have positive and significant coefficients of coresidence with grandparents. That is, mothers who live together with grandparents are 16.6% more likely to return to work earlier. Furthermore, we have positive and significant
15
First Draft coefficients of smaller firm size and negative and significant coefficients of firm size being public sector. That is, mothers who work for larger firms are less likely to return to work earlier. For the relationship between firm size and childcare leave, Ikeda (2010) analyze the factors which determine women’s job continuity. In columns (d), however, we do not have a significant coefficient of predicted mother’s early return to work. That is, we do not have causal impact of mothers’ early return to work after childbirth on their employment status as fulltime workers in 2004. Instead, we have positive and significant coefficient of mother’s university degree and a negative and significant coefficient of the number of siblings. In addition, we have negative and significant coefficients of smaller firm size and positive and significant coefficients of firm size being public sector, as well as a negative and significant coefficient of living in larger cities. In columns (e) and (f), we present the results of bivariate probit estimations where both the outcome variable Y and treatment variable T are assumed to be dummies as in equations (2). Here in column (e), we also have a positive and statistically significant coefficient of July and insignificant coefficient of mother’s early return to work, as in the case of 2SLS estimation.
2. The Results of Estimations for Mothers’ Working as Full‐time, Part‐time, Self‐ employed or Family worker, or Pieceworker in 2004 In Table 3 we show the results of OLS, probit, 2SLS and bivariate probit estimations of outcome variable Y (mother’s working as full-time, part-time, self-employed or family worker, or pieceworker at home) at 2004. In columns (a) and (b), our OLS and probit models suggest that a mother who returns to work early within 9 months after childbirth are 8.2-8.4% points more likely to work as full-time, part-time, self-employed or family worker, or pieceworker at home in 2004. In columns (d) and (f) of 2SLS and bivariate probit model we do not have a significant coefficient of predicted mother’s early return to work. That is, we do not have causal impact of mothers’ early return to work after childbirth on their employment status as full-time, part-time, self-employed or family worker, or pieceworker at home in 2004. 16
First Draft
3. The Results for Mothers’ Fulltime Working in 2011 Next in Table 4 and Table 5 we present the results of outcome variable in the year 2011. In Table 4, where the outcome variable Y is mother’s fulltime working in 2011 we have positive and significant coefficients of predicted treatment variable of early return to work in columns (d) of 2SLS and (f) of bivariate probit model. That is, if mothers return to work early within 9 month after childbirth, then they are causally 26.9%-33.6 points more likely to work as fulltime workers in 2011. In Table 5, however, we do not have significant coefficients of predicted treatment variable of early return to work in columns (d) of 2SLS and (f) of bivariate probit model. That is, even if mothers return to work early within 9 month after childbirth, then there is no causal impact on their work status in 2011 as full-time, part-time, self-employed or family worker, or pieceworker at home.
8. Discussion and Conclusion The results of our analyses show that mothers’ early return to work has causal positive impact on their employment status as fulltime workers in the long run not short run. The ground of the argument is as follows. First, our 2SLS and bivariate probit estimations show that mothers who return to work early after childbirth are more likely to work as fulltime workers in 2011 (columns (d) and (f) in Table 4). This shows a possibility that mother’s employment status as fulltime workers are causally affected by their early return to work after their child birth. On the other hand, when we change the outcome variable to mothers’ working as a fulltime, part-time, self-employed or family worker, or pieceworker at home, we do not have significant coefficient of the treatment variable, early return to work. That is, mother’s employment status as fulltime part-time, self17
First Draft employed or family worker, or pieceworker at home are not causally affected by their early return to work after their child birth in 2011 (Table 5). From these result we find that early return to work after their child birth has a causal positive effect on mother’s employment status as fulltime working, but not on the other working status such as part-time, self-employed or family worker, or pieceworker at home. As we stated in Introduction, no study in Japan have targeted policy intervention of childcare and parental leave to mothers’ return to work after childbirth in the context of quasi-experimental settings．However, Yamaguchi (2015) estimates a dynamic discrete choice structural model of in the presence of parental leave legislation and evaluate parental leave expansions that change the duration of job protection and/or the replacement rate of the cash benefit. He finds that a one-year job protection significantly increased maternal employment and earnings, but extending it from one to three years and offering cash benefits have little effect. His estimation results are consistent with our that mother’s employment status as fulltime workers are causally affected by their early return to work after their child birth, so it is the most convincing results. In Japan, little is known about which policy interventions have positive effects on mothers’ return to work after childbirth and job continuity, as Asai (2015), who conducted a difference-in-differences analysis focusing on an increase in replacement rate of childcare leave benefit from 25% to 40% in 2001, finds little evidence that job continuity increased in response to the reform. In addition, using a difference-in-differences analysis, Nagase (2014) also find that work continuation after the first childbirth was not affected by the 2008-2009 Japanese reform on short-hour option for employee with children under three, which was mandated for firms with 2001 employees. It is just possible that provision of more nursery facilities is an effective one because many studies point out the problem of wait-listed children waiting for existing childcare center insufficiency to accommodate children of working mothers (Lee and Lee (2014) and Zhou and Oishi (2005)).
18
First Draft
引用文献 Angrist, J. D., & Krueger, A. B. (1991). Does Compulsory School Attendance Affect Schooling and Earnings? The Quarterly Journal of Economics, 106(4), 979-1014. Asai, Y. (2015). Parental leave reforms and the employment of new mothers: Quasi-experimental evidence from Japan. Labour Economics, 36, 72–83. Baker, M., & Milligan, K. (2008). How Does Job-Protected Maternity Leave Affect Mothers’ Employment? Journal of Labor Economics, 26(4), 655-691. Bauernschuster, S., & Schlotter, M. (2015). Public child care and mothers' labor supply—Evidence from two quasiexperiments. Journal of Public Economics, 123, 1-16. Baum, C. L. (2003). The effect of state maternity leave legislation and the 1993 Family and Medical Leave Act on employment and wages. Labour Economics, 10(5), 573-596. Berlinski, S., Galiani, S., & McEwan, P. (2011). Preschool and Maternal Labor Market Outcomes: Evidence from a Regression Discontinuity Design. Economic Development and Cultural Change, 59(2), 313-344. Buckles, K. S., & Hungerman, D. M. (2013). Season of Birth and Later Outcomes: Old Questions, New Answers. Review of Economics and Statistics, 95(3), 711-724. Dickert-Conlin, S., & Chandra, A. (1999). Taxes and the Timing of Births. Journal of political Economy, 107(1), 161177. Gans, J. S., & Leigh, A. (2009). Born on the first of July: An (un) natural experiment in birth timing. Journal of Public Economics, 93(1), 246-263. Gelbach, J. (2002). Public Schooling for Young Children and Maternal Labor Supply. American Economic Review, 92(1), 307-322.
19
First Draft Geyer , J., Haan , P., & Wrohlich, K. (2014). The Effects of Family Policy on Mothers’ Labor Supply: Combining Evidence from a Structural Model and a Natural Experiment. SOEPpaper. Ikeda, S. (2010). Company Size and Childcare Leave: The Problems of Support for Women's Job Continuity. Japan Labor Review, 7(1), 119-139. Kureishi, W., & Wakabayashi, M. (2008). Taxing the stork. National Tax Journal, 167-187. Lalive, R., & Zweimüller, J. (2009). How does parental leave affect fertility and return to work? Evidence from two natural experiments. The Quarterly Journal of Economics, 124(3), 1363-1402. Lalive, R., Schlosser, A., Steinhauer, A., & Zweimüller, J. (2014). Parental leave and mothers' careers: The relative importance of job protection and cash benefits. The Review of Economic Studies, 81(1), 219-265. Lee, G. H., & Lee , S. (2014). Childcare availability, fertility and female labor force participation in Japan. Journal of the Japanese and International Economies, 32, 71–85. Matsuda, S., & Kahyo, H. (1992). Seasonality of Preterm Births in Japan. International Journalof Epidemiology, 21(1), 91-100. Matsuda, S., & Kahyo, H. (1998). Geographic Differences in Seasonality of Preterm Births in Japan. Human Biology, 70(5), 919-935. Ministry of Health, Labour and Welfare. (2002). The Basic Survey of Women Workers' Employment Management. Nagase, N. (2014). The Effect of a Short-Hour Option Mandate on First Childbirth, Birth Intention and Work Continuation after Childbirth : Evidence from Japan. The Journal of Population Studies, 37(1), 29-53. Sakata, A., Tano, J., & Fuse, K. (2015). For Ministry of Health, Labour and Welfare longitudinal survey (development and challenges of the Special Panel Survey). Advances in Social Research, 15, 21-29. Schönberg, U., & Ludsteck, J. (2007). Maternity leave legislation, female labor supply, and the family wage gap. IZA discussion paper.
20
First Draft Yamaguchi , S. (2015). Effects of Parental Leave Policies on Female Career and Fertility Choices. Available at SSRN. Yamaguchi , S. (2015). Family Policies and Female Employment in Japan. Available at SSRN. Zhou, Y., & Oishi, A. (2005). Underlying Demand for Licensed Childcare Services in Urban Japan. Asian Economic Journal, 19(1), 103–119.
21
First Draft
Tables
Table 1 (a): Descriptive statistics (year 2002)
Mean Dependent Variables mother's fulltime working 0.827 mother's working 0.872 Independent Variables early return to work (8 months) 0.452 early return to work (9 months) 0.492 mother's months of childcare leave 7.508 Instrumental Variable July 0.498 valid job vacancy rate 0.987 father's annual income 2000* 429.600 father's log annual income 2000 5.971 Control Variables mother's age 30.635 mother's university degree 0.238 number of siblings 0.069 father's annual income (last year)* 442.033 father's log annual income (last year) 5.993 n=3021
2002 Std. Dev. Min
Max
0.378 0.335
0 0
1 1
0.498 0.500 4.405
0 0 0
1 1 12
0.500 0.217 197.496 0.455
0 0.49 16 2.773
1 1.49 5000 8.517
3.945 0.426 0.257 197.629 0.477
18 0 0 13 2.565
46 1 2 2500 7.824
Source: the Longitudinal Survey of Newborns in the 21st Century, wave 2002. n= 3,021 Sample selection: 1) babies are first-born,2) full-time workers both one year before delivery and a half year after delivery, 3) answer other questions providing necessary information. Note: * 10,000 yen
22
First Draft
Figure 1 (b): Descriptive statistics (year 2011)
Mean Dependent Variables mother's fulltime working 0.624 mother's working 0.863 Independent Variables early return to work (8 months) 0.427 early return to work (9 months) 0.465 mother's months of childcare leave 7.943 Instrumental Variable July 0.494 valid job vacancy rate 0.995 father's annual income 2000* 441.687 father's log annual income 2000 6.000 Control Variables mother's age 39.785 mother's university degree 0.258 number of siblings 1.047 father's annual income (last year)* 537.798 father's log annual income (last year) 6.179 n=2387
2011 Std. Dev. Min
Max
0.485 0.343
0 0
1 1
0.495 0.499 4.986
0 0 0
1 1 36
0.500 0.217 202.489 0.455
0 0.49 18 2.890
1 1.49 5000 8.517
3.897 0.438 0.697 269.699 0.498
27 0 0 3 1.099
55 1 4 5300 8.575
Source: the Longitudinal Survey of Newborns in the 21st Century, wave 2011. n= 2,387 Sample selection: 1) babies are first-born,2) full-time workers both one year before delivery and a half year after delivery, 3) answer other questions providing necessary information. Note: * 10,000 yen
23
First Draft Table 2: Estimation Results of Mother’s Fulltime Working in Year 2004 Outcome Variable: Y OLS Coefficient (a) Treatment Variable: T early return to work
0.1119*** [0.017]
mother's fulltime working at 2004 2SLS 1st stage 2nd stage Marginal Effects Coefficient Coefficient (b) (c) (d) Probit
0.1144*** [0.017] 0.0765*** [0.018] ‐0.2147*** [0.059]
valid job vacancy rate
mother's age^2 mother's university degree number of siblings father's log annual income coresidence large cities other cities firm size (1~4) firm size (5~99) firm size (100~499) firm size (public sector) Constant
0.0192 [0.024] ‐0.0002 [0.000] 0.0621*** [0.018] ‐0.1019*** [0.015] ‐0.0063 [0.005] 0.0489*** [0.018] ‐0.0896*** [0.027] ‐0.0229 [0.021] ‐0.1867*** [0.063] ‐0.0577** [0.023] 0.0009 [0.023] 0.2222*** [0.020] 0.4336 [0.410]
Coefficient (e)
Coefficient (f)
0.2428 [0.158]
Instruments: Z July
Control variables: mother's age
Bivariate Probit
0.0073*** [0.002]
0.2161*** [0.048] ‐0.2687* [0.155]
‐0.0495* [0.026] 0.0007* [0.000] ‐0.0548** [0.022] ‐0.0163 [0.017] ‐0.0160*** [0.006] 0.1656*** [0.020] 0.0115 [0.030] 0.0166 [0.023] 0.3108*** [0.057] 0.1953*** [0.024] 0.0726*** [0.025] ‐0.1057*** [0.026] 1.5799*** [0.435]
0.0666*** [0.019] ‐0.0898*** [0.015] ‐0.0787*** [0.021]
‐0.1919*** [0.067] ‐0.0707*** [0.023] ‐0.0059 [0.022] 0.2230*** [0.016] 1.3252*** [0.428]
0.538 [0.409]
0.0269 [0.025] ‐0.0003 [0.000] 0.0692*** [0.021] ‐0.0998*** [0.016] ‐0.0042 [0.006] 0.0278 [0.032] ‐0.0907*** [0.027] ‐0.0256 [0.021] ‐0.2260*** [0.073] ‐0.0838** [0.038] ‐0.0087 [0.025] 0.2361*** [0.029] 0.2171 [0.473]
0.0008 [0.007]
0.0236*** [0.007]
‐0.1210** [0.059] ‐0.0064 [0.046] ‐0.1740** [0.081]
0.2323*** [0.070] ‐0.2897*** [0.052] ‐0.2400*** [0.087]
0.8692*** [0.189] 0.4958*** [0.069] 0.1807*** [0.070] ‐0.3086*** [0.078] 1.5425*** [0.424]
‐0.5891*** [0.209] ‐0.2518** [0.099] ‐0.0305 [0.074] 0.9300*** 1.1606* [0.656] ‐0.1029 [1.380] 0.3248 [0.119]
athrho Estimated Joint Probabilities biprob11 Observations R‐squared Adjusted R‐Squared F‐statistic Log likelihood Pseudo R‐squared LR Sargan statistic P‐value of Sargan statistic Standard errors in brackets *** p<0.01, ** p<0.05, * p<0.1
2,958 0.087 0.0833 29.05
2,958
2,937 0.068 0.0635 17.88 ‐1540 0.0883 298.3 0.13 0.719
24
2,937 0.105 0.101 30.37
First Draft Table 3: Estimation Results of Mother’s Working (as a full-time, part-time, self-employed or family worker, or pieceworker at home) in Year 2004
Output Variable: Y
Treatment Variable: T early return to work
mother's working (as a full‐time, part‐time, self‐employed or family worker, or pieceworker at home) at 2004 2SLS Bivariate Probit OLS Probit Coefficient Marginal Effects Coefficient Coefficient Coefficient Coefficient (a) (b) (c) (d) (e) (f) 0.0822*** [0.015]
0.0837*** [0.013]
0.0498 [0.135]
Instruments: Z July
0.0765*** [0.018] ‐0.2147*** [0.059]
valid job vacancy rate Control variables: mother's age mother's age^2 mother's university degree number of siblings father's log annual income coresidence large cities other cities firm size (1~4) firm size (5~99) firm size (100~499) firm size (public sector) Constant
0.0058 [0.021] ‐0.0001 [0.000] 0.0584*** [0.015] ‐0.1278*** [0.013] ‐0.0047 [0.005] 0.0597*** [0.015] ‐0.0586** [0.024] ‐0.0029 [0.018] ‐0.0444 [0.052] 0.0276 [0.020] 0.0148 [0.021] 0.1765*** [0.018] 0.7851** [0.361]
0.0021 [0.002]
0.2147*** [0.049] ‐0.2656* [0.148]
‐0.0495* [0.026] 0.0007* [0.000] ‐0.0548** [0.022] ‐0.0163 [0.017] ‐0.0160*** [0.006] 0.1656*** [0.020] 0.0115 [0.030] 0.0166 [0.023] 0.3108*** [0.057] 0.1953*** [0.024] 0.0726*** [0.025] ‐0.1057*** [0.026] 1.5799*** [0.435]
0.0593*** [0.015] ‐0.1066*** [0.012] ‐0.0710*** [0.017]
‐0.0459 [0.052] 0.0126 [0.017] 0.0068 [0.017] 0.1555*** [0.012] 2.3526*** [0.483]
0.4298 [0.310]
0.0039 [0.021] ‐0.0001 [0.000] 0.0567*** [0.018] ‐0.1283*** [0.013] ‐0.0053 [0.005] 0.0649** [0.027] ‐0.0583** [0.023] ‐0.0023 [0.018] ‐0.0346 [0.063] 0.034 [0.033] 0.0172 [0.021] 0.1731*** [0.025] 0.8387** [0.404]
0.0009 [0.007]
0.0093 [0.008]
‐0.1217** [0.059] ‐0.0067 [0.047] ‐0.1685** [0.076]
0.2846*** [0.073] ‐0.4685*** [0.059] ‐0.3072*** [0.078]
0.8683*** [0.193] 0.4958*** [0.066] 0.1804*** [0.066] ‐0.3074*** [0.073] 1.5515*** [0.411]
‐0.2029 [0.218] 0.045 [0.102] 0.0262 [0.080] 0.9402*** [0.109] 2.2955*** [0.553] ‐0.0359 [0.211] 0.3687 [0.1450]
athrho Estimated Joint Probabilities biprob11 Observations R‐squared Adjusted R‐Squared F‐statistic Log likelihood Pseudo R‐squared LR Sargan statistic P‐value of Sargan statistic Standard errors in brackets *** p<0.01, ** p<0.05, * p<0.1
2,958 0.08 0.0762 22.01
2,958
2,937
2,937 0.079 0.0745 17.13
‐1216 0.0967 260.4 0.0189 0.891
25
First Draft Table 4: Estimation Results of Mother’s Fulltime Working in Year 2011 Outcome Variable: Y OLS Coefficient (a) Treatment Variable: T early return to work
0.1114*** [0.020]
mother's fulltime working at 2011 2SLS 1st stage 2nd stage Marginal Effects Coefficient Coefficient (b) (c) (d) Probit
0.1258*** [0.021] 0.0839*** [0.020] ‐0.1893*** [0.064]
valid job vacancy rate
mother's age^2 mother's university degree number of siblings father's log annual income coresidence large cities other cities firm size (1~4) firm size (5~99) firm size (100~499) firm size (public sector) Constant
0.0047 [0.035] 0 [0.000] 0.0811*** [0.022] ‐0.0216 [0.014] ‐0.0171*** [0.005] 0.1091*** [0.021] ‐0.0884** [0.036] ‐0.0416 [0.031] ‐0.1613** [0.072] ‐0.0656** [0.027] ‐0.0186 [0.028] 0.2803*** [0.025] 0.6137 [0.713]
Coefficient (e)
Coefficient (f)
0.3363* [0.191]
Instruments: Z July
Control variables: mother's age
Bivariate Probit
0.0087*** [0.003]
0.2166*** [0.064] ‐0.1734** [0.076]
‐0.0667* [0.035] 0.0008* [0.000] ‐0.0624*** [0.024] 0.0083 [0.015] ‐0.0055 [0.006] 0.1191*** [0.023] ‐0.043 [0.038] ‐0.03 [0.033] 0.2813*** [0.070] 0.1822*** [0.027] 0.0786*** [0.029] ‐0.1184*** [0.028] 1.9898*** [0.718]
0.0774*** [0.024] ‐0.0185 [0.016] ‐0.0731*** [0.022]
‐0.1538* [0.082] ‐0.0802*** [0.028] ‐0.0239 [0.028] 0.2975*** [0.022]
1.1060** [0.457]
0.0221 [0.038] ‐0.0002 [0.000] 0.0949*** [0.026] ‐0.0232 [0.015] ‐0.0155** [0.006] 0.0840*** [0.031] ‐0.0766** [0.038] ‐0.0335 [0.033] ‐0.2218** [0.089] ‐0.1067** [0.044] ‐0.0358 [0.031] 0.3078*** [0.037] 0.1246 [0.826]
0.0028 [0.008]
0.0225*** [0.007]
‐0.1179* [0.067] 0.0281 [0.043] ‐0.0621 [0.068]
0.2348*** [0.067] ‐0.0524 [0.041] ‐0.1363 [0.089]
0.6726*** [0.217] 0.4732*** [0.075] 0.1803** [0.076] ‐0.3552*** [0.083] 1.3576*** [0.506]
‐0.5703*** [0.212] ‐0.3400*** [0.102] ‐0.1141 [0.076] 0.9607*** [0.111] ‐0.3257 [0.715] ‐0.554 [4.440] 0.269 [0.095]
athrho Estimated Joint Probabilities biprob11 Observations R‐squared Adjusted R‐Squared F‐statistic Log likelihood Pseudo R‐squared LR Sargan statistic P‐value of Sargan statistic Standard errors in brackets *** p<0.01, ** p<0.05, * p<0.1
2,411 0.105 0.0997 30.72
2,411
2,387 0.055 0.0501 18.28 ‐1448 0.0837 264.6 0.0237 0.878
26
2,387 0.094 0.0883 21.12
First Draft Table 5: Estimation Results of Mother’s Working (as a full-time, part-time, self-employed or family worker, or pieceworker at home) in Year 2011
Output Variable: Y
Treatment Variable: T early return to work
mother's working (as a full‐time, part‐time, self‐employed or family worker, or pieceworker at home) at 2011 2SLS Bivariate Probit OLS Probit Coefficient Marginal Effects Coefficient Coefficient Coefficient Coefficient (a) (b) (c) (d) (e) (f) 0.0500*** [0.015]
0.0539*** [0.014]
‐0.0902 [0.138]
Instruments: Z July
0.0839*** [0.020] 1.9898*** [0.718]
valid job vacancy rate Control variables: mother's age mother's age^2 mother's university degree number of siblings father's log annual income coresidence large cities other cities firm size (1~4) firm size (5~99) firm size (100~499) firm size (public sector) Constant
0.0393 [0.027] ‐0.0005 [0.000] 0.0139 [0.016] ‐0.0380*** [0.011] ‐0.0099*** [0.004] 0.0595*** [0.014] ‐0.0267 [0.027] 0.0051 [0.023] ‐0.096 [0.066] 0.0392* [0.020] 0.0245 [0.021] 0.1276*** [0.019] 0.238 [0.535]
0 [0.002]
‐0.0624*** [0.024] 0.0083 [0.015] ‐0.0624*** [0.024] 0.0083 [0.015] ‐0.0055 [0.006] 0.1191*** [0.023] ‐0.043 [0.038] ‐0.03 [0.033] 0.2813*** [0.070] 0.1822*** [0.027] 0.0786*** [0.029] ‐0.1184*** [0.028] ‐0.1893*** [0.064]
0.0135 [0.016] ‐0.0320*** [0.010] ‐0.0806*** [0.015]
‐0.1482** [0.073] 0.0107 [0.017] 0.0137 [0.017] 0.1108*** [0.013] 3.3131*** [0.586]
0.6325 [0.401] 0.2202*** [0.057] ‐0.4687** [0.182]
0.0285 [0.027] ‐0.0004 [0.000] 0.0053 [0.019] ‐0.0370*** [0.011] ‐0.0109** [0.004] 0.0751*** [0.022] ‐0.0341 [0.028] 0.0001 [0.024] ‐0.0583 [0.064] 0.0648** [0.032] 0.0352 [0.022] 0.1105*** [0.027] 0.5429 [0.597]
0.0027 [0.007]
0.0005 [0.009]
‐0.1173* [0.065] 0.0269 [0.041] ‐0.0489 [0.067]
0.0856 [0.085] ‐0.1617*** [0.052] ‐0.3787*** [0.124]
0.6863*** [0.219] 0.4701*** [0.075] 0.1780** [0.077] ‐0.3529*** [0.086] 1.3928*** [0.504]
‐0.6487*** [0.243] ‐0.0121 [0.111] 0.0443 [0.095] 0.7482*** [0.119] 2.9634*** [0.909] ‐0.2302 [0.279] 0.3727 [0.145]
athrho Estimated Joint Probabilities biprob11 Observations R‐squared Adjusted R‐Squared F‐statistic Log likelihood Pseudo R‐squared LR Sargan statistic P‐value of Sargan statistic Standard errors in brackets *** p<0.01, ** p<0.05, * p<0.1
2,411 0.042 0.037 8.859
2,411 0.094 0.079 0.0883 0.0745 21.12 17.13
2,387
‐884.4 0.0707 134.6 0.0115 0.914
27
2,937
First Draft
Figures
Figure 1: The Ratio of Day Care Center Utilization
0.50 0.45 0.40 0.35
Ratio
0.30 0.25 0.20 0.15 0.10 0.05 0.00 2002
2003
2004
2005
2006
2007
2008
2009
2010
2011
2012
2013
2014
year 0 year old
1 & 2 year old
3 year old
4 or more year old
Source: Authors' calculation using data from the Ministry of Health, Labour and Welfare’s Report on Social Welfare Administration and Services, and Population Estimates Note: The utilization ratio of child care facilities is defined as the number of children being admitted in child care facilities in the age group/ the number of children in the age group.
28
First Draft Figure 2 (a): Mothers' Employment Status at 2004
Early return to work = 1 (n=1,531)
Early return to work = 0 (n=1,683)
0
10
20
30
40
50
60
70
80
90
100
% Full‐time
Part‐time
Selfemploy
Domestic piecework
Housewives, No work, Students
(b): Mothers' Employment Status at 2011
Early return to work = 1 (n=1,250)
Early return to work = 0 (n=1,408)
0
10
20
30
40
50
60
70
80
90
100
% Full‐time
Part‐time
Selfemploy
Domestic piecework
29
Housewives, No work, Students
First Draft
Figure 3: The Period of Childcare Leave
30 25
%
20 15 10 5 0 0
1
2
3
4
5
6
7
8
9
10
11
12
13 or more
months January
July
Source: the Longitudinal Survey of Newborns in the 21st Century, wave 2001. n= 3,151 Note: The question explicitly instructed respondents to exclude maternity leave and short-term working.
30
First Draft
Figure 4
0.3
0.5
0.2
0.4
0.1
0.3
0
0.2 January
July
January
Mother' University Degree
July
Father's University Degree
33 32.5 32 31.5 31 30.5 30
33 32.5 32 31.5 31 30.5 30 January
July
January
Mother's age
July
Father's age
33 32.5 32 31.5 31 30.5 30
33 32.5 32 31.5 31 30.5 30 January
July
January
Mother's age
Father's age
31
July
First Draft
460 450 440 430 420 410
33 32.5 32 31.5 31 30.5 30 January
July
January
Father's Annual Income 2000
July
Mother's age
33 32.5 32 31.5 31 30.5 30
460 450 440 430 420 410 January
July
January
Father's age
July
Father's Annual Income 2000
32
