Master Thesis in Economics, 30 credits
Master of Science in Business and Economics, 240 credits
Spring term 2021
The Family Gaps in Pay
Empirical Evidence From Japan
I would like to thank my supervisor, Professor Giovanni Forchini, for the valuable guidance, support and kind advice throughout the construction of this thesis.
Table of Contents
1. Introduction ... 1
1.1 Background ... 1
1.2 Aims and Objectives ... 3
1.3 Outline of the Thesis ... 4
2. Literature Review ... 5
2.1 Explaining the Family Pay Gap ... 5
2.1.1 The Motherhood Wage Penalty ... 5
2.1.1 The Fatherhood Wage Premium ... 8
2.1.3 Overview of the Theoretical Framework ... 10
2.2 Empirical Evidence ... 11
2.3 Heterogeneity of Pay Gaps ... 13
2.3.1 The Family Pay Gap by Educational Attainment ... 13
2.3.2 The Family Pay Gap by Age Groups ... 14
2.3.3 Heterogeneity of Pay Gaps in Japan ... 15
2.4 Work Culture, Family Policy and Gender Equality in Japan ... 16
3. Methodology ... 20
3.1 Japan Household Panel Survey ... 20
3.2 Research Design ... 21
3.3 Creation of the Panel Data Set ... 24
4. Analysis and Discussion ... 28
4.1 Descriptive Statistics ... 28
4.2 Main Analysis ... 30
4.2.1 The Full Sample of Women ... 30
4.2.2 The Full Sample of Men ... 33
4.2.3 Summary Discussion of the Main Analysis ... 36
4.3 Further Analysis of the Motherhood Wage Penalty ... 38
4.3.1 Subsamples by Educational Attainment ... 38
4.3.2 Subsamples by Age Group ... 40
5. Conclusions, Limitations and Recommendations ... 43
5.1 Conclusion ... 43
5.2 Limitations ... 44
5.3 Suggestions for Policy Measures and Future Research ... 46
Appendix A: Descriptive Statistics ... 56
Appendix B: Econometric Tests ... 61
Appendix C: Regression Results ... 63
List of FiguresFigure 1: Gender Gap Index for Japan………1
Figure 2: The Increased Role of Parenthood in the Gender Earnings Gap………..…...2
Figure 3: Summary of the Main Explanations Behind the Family Wage Gaps………10
Figure 4: The Magnitude of the Japanese Family Pay Gaps………...36
Figure 5: The Motherhood Wage Penalty by Educational Attainment……….39
Figure 6: The Motherhood Wage Penalty by Age Groups………..…………...41
List of TablesTable 1: The Main Analysis of Women.………...….……31
Table 2: The Main Analysis of Men………...….…..34
Table A1: Descriptive Statistics of the Sample of Women……….……..56
Table A2: Descriptive Statistics of the Sample of Men……….….…...58
Table B1: Breusch-Pagan Test for Heteroskedasticity in Pooled OLS Models……….……...61
Table B2: Modified Wald Test for Groupwise Heteroskedasticity in Fixed Effects Models....61
Table B3: Robust Hausman Test for Fixed Effects vs. Random Effects……….…….….62
Table C1: Regression Results From the Main Analysis of Women……….….…63
Table C2: Regression Results From the Main Analysis of Men……….….….…64
Table C3: Fixed Effects Results From Subsamples by Educational Attainment……….….…66
Japan has the second largest gender wage gap among the OECD nations (OECD, 2021) and is currently ranked 120 out of 156 countries worldwide in terms of gender equality, making it one of the lowest ranked countries in the East Asia and Pacific region (World Economic Forum, 2021). Despite the Japanese government’s effort to narrow the gender-based gaps by promoting women’s economic participation, the progress towards closing them is slow relative to the rest of the world (Yamaguchi, 2019). Figure 1 illustrates the slow-moving distance to parity measured by score as Japan is falling behind in the progress of gender equality. OECD (2017b) estimates that due to maintained traditional gender roles, Japanese women still do more than three-quarters of the unpaid work and caregiving while men work long office hours. At the same time, Japan is trying to overcome a persistently low fertility rate which has declined significantly over the past decades (SBJ, 2020, p. 16). With an old and rapidly shrinking population, the diminishing labor force is expected to reduce Japan’s real gross domestic product by over 25 percent in the next four decades if no reforms are implemented (Colacelli & Corugedo, 2018). One reason for the low fertility rates is that Japanese women struggle to balance work and family life, causing them to have fewer children as a result of the work-family conflict (e.g. Boling, 2008; Hertog & Kan, 2017; Nae, 2018).According to the national research institute IPSS (2017, p. 17), approximately 47% of Japanese women choose to leave their place of employment after having their first child.
79 91 98 101 94 98 101 105 104 101 111 114 110 121 120 0.645 0.645 0.643 0.645 0.652 0.651 0.653 0.650 0.658 0.670 0.660 0.657 0.662 0.652 0.656 0.550 0.570 0.590 0.610 0.630 0.650 0.670 0.690 30 40 50 60 70 80 90 100 110 120 130 2006 07 08 09 10 11 12 13 14 15 16 17 18 20 21 Sco re Ra nk
Gender Gap Index for Japan
2 There is a growing literature indicating that parenthood greatly contributes to the gender wage gap. Trends in women’s earnings indicate that mothers have experienced the slowest wage growth over time, causing working mothers to account for the majority of the pay gap between men and women (Glass, 2004). Kleven, Landais & Søgaard (2019) show that the proportion of gender disparity attributable to parenthood has increased dramatically over time, from 40% in 1980 to approximately 80% just over three decades later. They further conclude that while the total earnings gap has narrowed, the child-related earnings gap has instead widened. The increased role of parenthood in the gender inequality in earnings is illustrated in Figure 2. Today, many studies provide evidence of the labor market consequences of motherhood, where particular attention is paid to the disadvantages in pay between mothers and childless women (Cukrowska-Torzewska & Matysiak, 2020). This phenomenon is often referred to as the motherhood wage penalty or child penalty and will be the centerpiece of this study. In contrast to the negative impact of children on women’s earnings, wage differences between men with and without children often show an opposite relationship (Meurs & Ponthieux, 2015). The wage bonus enjoyed by fathers is commonly known as the fatherhood wage premium. Although some studies find men’s wages to be unaffected by parenthood, a gender wage gap is in general always wider among parents than among nonparents (Grimshaw & Rubery, 2015).
Figure 2: The Increased Role of Parenthood in the Gender Earnings Gap. Source: Kleven, Landais & Søgaard (2019, p. 199).
3 has been almost no research into the motherhood wage penalty in Japan and the impact of children on wages is not well established. Takeuchi (2018) performs a cross-sectional analysis without the ability to control for time-invariant heterogeneity and argues that verification of the wage penalty is needed with various data. The presence of a fatherhood wage premium has also been scarcely investigated as Yukawa (2015) provides the only evidence of the phenomena within a Japanese context. Because of the limited research regarding Japanese family pay gaps in combination with Japan’s substantial wage inequalities between genders, further research is needed to investigate how parenthood affects the wages of Japanese men and women. If, as research suggests, children play a significant role in the wage disparity between genders, understanding how and why parenthood influences wages is necessary to combat the wide gender gap in pay. Further, if women’s labor supply as well as fertility decisions are greatly affected by the work-family conflict, facilitating Japanese women’s possibility to invest in both a career and a family by increased gender equality is of great importance. Grimshaw & Rubery (2015) argue that the existence of a wage penalty for motherhood raises concerns not only for gender equality, but also for the society’s ability to strike a long-term balance between the economic interests of female labor force participation and the goals of ensuring an equal distribution of income to support reproduction and childrearing.
1.2 Aims and Objectives
The main aim of this thesis is to quantify the adjusted family pay gaps1 in the Japanese labor
market by investigating the impact of dependent children on wages. In this study, dependent children are defined as children under the age of 18 living in the household. Because most researchers restrict their analysis to the pay gap between mothers and childless women, an analysis of how the wages of Japanese men are affected by parenthood is also included. The presence of one, two or three or more dependent children helps to determine whether a wage gap varies nonlinearly and if it widens with more children. This study further attempts to explain which factors mainly contribute to the presence of a motherhood wage penalty or a fatherhood wage premium, which has not yet been sufficiently investigated in a Japanese context. At present, no longitudinal study has tried to explain the relative importance of different factors in shaping the Japanese family pay gaps.
1 The unadjusted family pay gaps measure the average difference in pay between parents and nonparents. The
4 Moreover, since children are mainly expected to have negative wage consequences for women, particular attention will be paid to this group as various subsamples are analysed to further identify additional sources of the pay gap between mothers and nonmothers. Specifically, analyses of whether a child penalty varies by educational attainment or by age groups of women are performed. The results from previous literature on how the motherhood pay gap varies by educational level differ and are somewhat contradictory. Furthermore, no analysis of whether the magnitude of the Japanese motherhood wage penalty varies by academic achievement has been performed previously. Investigating the motherhood pay gap by age groups further helps to determine whether the wages of mothers eventually catch up to those of nonmothers or if the pay gap widens as they grow older. This is an important question because persistent child penalties could lead to a considerable pay gap over time and leave mothers with a substantially lower life income and poorer pension.
Given the need to improve the understanding of how parenthood influences wages of Japanese women and men, this study seeks to answer the following research questions: (1) What are the magnitudes of the Japanese family pay gaps and which factors mainly contribute to their existence? (2) Does educational attainment affect the size of the motherhood pay gap such that highly educated women experience the largest motherhood wage penalty? (3) Do the wages of mothers eventually catch up to those of childless women over time, or do they fall further behind as they age?
1.3 Outline of the Thesis
2. Literature Review
This chapter starts by presenting the theoretical framework that seeks to explain the family pay gaps. The main economics-based theories are introduced together with some of the empirical evidence linked to the specific theories. Thereafter, additional evidence of family wage gaps from previous literature is reviewed, including the existing papers from Japan. In Chapter 2.3, a discussion on how pay gaps differ by characteristics of the parents is presented where particular focus is paid to variations in the motherhood wage penalty by educational attainment and age groups. The heterogeneity of pay gaps from the existing literature from Japan is also reviewed. Lastly, a brief overview of work culture, family policy and gender equality in Japan is given to provide additional background on the Japanese context.
2.1 Explaining the Family Pay Gap
2.1.1 The Motherhood Wage Penalty
6 out of the labor market, Staff & Mortimer (2012) conclude that this factor fully explains the residual pay gap between mothers and nonmothers among young women in the U.S.
7 Further, it is also argued that a possible explanation behind the wage penalty is due to disparities in work effort between mothers and nonmothers. According to this hypothesis, mothers invest less commitment into work because they are involved in childcare and housework and have less energy to exert at work (Becker, 1985). This explanation thus states that the wage penalty may be due to actual productivity differences. The work effort hypothesis is rarely tested in empirical research because of the difficulty of finding good measures of work effort such as productivity. Kmec (2011) uses self-reported measures and does not find any evidence that mothers would exert less work effort or have less motivation to work because of their family. Contrarily, Azmat & Ferrer (2017) show that having young children actually reduces women’s job performance among lawyers, which could further decrease the probability of being promoted. In their study, hours billed to clients and new client revenue raised are used as measures of productivity. Although work effort is difficult to observe, Cukrowska-Torzewska & Matysiak (2020) argue that the work effort hypothesis could be tested by performing analyses on subsamples distinguished by the level of education or age. This is further discussed in Chapter 2.3.
8 Lastly, selection into motherhood is a factor that is often considered when explaining the wage inequalities between mothers and nonmothers (Gough & Noonan, 2013). This suggests that women who choose to have children may differ from childless women with characteristics that also affect their wages.For instance, women who decide to have many children may be more family-oriented and focus less on career aspirations. Gough & Noonan (2013) argue that before childbirth, women who become mothers may differ in motivation and work commitment compared to nonmothers. Mincer & Polachek (1974) also discuss that women who plan to have children may invest less in human capital activities than those who do not. Other characteristics that could differ between mothers and nonmothers include aspects of life cycle plans, future orientation and attitudes or preferences about work and family (Budig & England, 2001). If for instance career aspirations are unchanged over time, this kind of often unmeasured and time-invariant heterogeneity could be controlled for by a model which eliminates such characteristics. Thus, selection into motherhood is often addressed with models based on panel data where fixed effects models are commonly used (Cukrowska-Torzewska & Matysiak, 2020). As an example, Anderson, Blinder & Krause (2002) show that controlling for time-invariant characteristics causes the motherhood wage penalty to fall dramatically. On the contrary, Dumauli (2019a) shows that the Japanese motherhood pay gap is greater with a fixed effects model compared to a pooled OLS model, which points to a positive selection into motherhood. This would imply that women whose time-invariant characteristics are associated with higher wages also somewhat predict greater fertility.
2.1.1 The Fatherhood Wage Premium
9 without children, fathers were also given a higher starting wage. However, discriminatory behaviors are as mentioned difficult to identify and there is no conclusive evidence of fathers being favored over childless men. In a relatively recent study from Sweden, Bygren, Erlandsson & Gähler (2017) find no evidence of discrimination on the basis of gender and parental status. Due to couple specialization, fathers may increase their work effort conditional on their partner devoting more time to the household (Mari, 2019). As women increase their investment in housework and childcare, men tend to spend more time on paid work which could increase wages over time. It has been shown that married men, specifically those whose wives do less paid work after entering parenthood, spend more hours working and gain wage premiums (Lundberg & Rose, 2000). Budig & Hodges (2010) show that the adjusted wage premium for fathers in the United States falls from 11% to 6% when controlling for marital status, which suggests that marriage helps to explain the fatherhood wage premium. Kmec (2011) also finds that fathers, more often than childless men, report that providing for their home makes them work harder which would generate higher wages. Yet, some studies show that compared to being childless, fathers do not spend less time onunpaid work which does not support the theory of household specialization (Killewald, 2012). Further, Killewald (2012) argues that fatherhood can cause men to change their labor market behavior in several ways that promote higher wages. For example, fatherhood may increase the feeling of responsibility and motivate fathers to prosocial and responsible behavior. Entering parenthood could also cause some fathers to move to higher-paying jobs or invest more in human capital rewarded in the labor market. Cooke & Fuller (2018) argue that under a male breadwinner norm, having children may motivate men to work longer, harder and change occupations to receive higher wages. Many fathers could also feel a greater moral imperative and a social pressure to provide financial support for their families (Percheski & Wildeman, 2008). Moreover, fatherhood has been found to be associated with greater job tenure, which likely contributes to higher wages (Millimet, 2000).
10 (Budig & Hodges, 2010). Not taking such factors into account would thus underestimate the fatherhood wage premium. By contrast, the opposite effect has been detected for some European countries which suggests that fathers are instead positively selected into fatherhood (Cooke & Fuller, 2018). For example, Kunze (2020) concludes that there is positive selection into fatherhood in Norway. Positive selection may for instance occur if fathers have unmeasured characteristics such as commitment and loyalty, which are positively valued by both employers and potential partners (Cooke & Fuller, 2018). Hence, characteristics that are associated with a greater likelihood of having children could be the same factors that contribute to higher wages (Budig & Hodges, 2010). As when studying the family pay gap among women, selection into fatherhood is often controlled for by using panel data models such as fixed effects models.
2.1.3 Overview of the Theoretical Framework
Figure 3 provides an overview of the main economics-based explanations of the motherhood wage penalty and the fatherhood wage premium from the literature and previous research.
Figure 3: Summary of the Main Explanations Behind the Family Wage Gaps.
The motherhood wage penalty The fatherhood wage premium
capital theory - Depreciation of knowledge and foregone work experience Household specialization - Reallocation of paid and unpaid work within couples
specialization - Reallocation of paid and unpaid work within couples Work effort hypothesis - Increased work effort and time spent on paid work
differentials - Selection into jobs with family-friendly arrangements Labor market discrimination - Discriminatory behaviors against childless men
hypothesis - Less commitment into work and reduced productivity Selection into fatherhood - Factors that affect wages are also associated with entering fatherhood
discrimination - Taste-based or statistical discrimination against mothers
2.2 Empirical Evidence
While the presence of a motherhood wage penalty appears to be universal, the magnitude of the wage penalty varies by countries and different regions of the world. For instance, the unadjusted pay gaps between mothers and nonmothers tend to be greater in developing countries (Grimshaw & Rubery, 2015).Most of the research into the family pay gap among women is based on U.S. data, although many studies have also been done in the United Kingdom along with other industrialized countries, especially in Europe (Meurs & Ponthieux, 2015). Hence, research has generally been less extensive in areas such as Asia or Africa. One of the early studies which received much attention was conducted by Waldfogel (1997), who show that an unexplained family wage gap causes American women with one child to experience a wage penalty of approximately 4%, whereas women with two or more children face a wage penalty closer to 12%. Other influential U.S. research include studies by for instance Budig & England (2001) and Anderson, Blinder & Krause (2003), who find evidence of adjusted wage penalties of around 3-5% per child. In recent meta-regression analyses, de Linde Leonard & Stanley (2020) estimate the wage penalty to be about 2-6% per child in the U.S., consistent with the estimate of Cukrowska-Torzewska & Matysiak (2020) who conclude that the unexplained wage gap amounts to 3-6%. Within Europe, Davies & Pierre (2005) make use of several data sets and find that out of 11 countries analysed, significant wage gaps exist in the United Kingdom, Ireland, Germany, Spain, Portugal and Denmark. They further show that the presence of one child is associated with a wage penalty of 2-6%, while the presence of two children causes wage rates to decrease by 7-12%. Empirical evidence often finds that the wage penalty for motherhood increases with the number of children (Grimshaw & Rubery, 2015).Therefore, women with one child may only face a small wage penalty while the presence of several children causes a sizeable pay gap. Statistically significant child penalties have regularly been found to be around 2-10% for one child and 5-15% for several children (Meurs & Ponthieux, 2015).
12 & Misra (2012) covers 22 industrialized countries, where statistically significant motherhood wage penalties are found in 13 of them after controlling for various characteristics.
Considering the impact of children on men’s wages, a wage premium is often relatively small in size whenever detected (Mari, 2019). In Germany, Pollmann-Schult (2011) find wage premiums for two or three or more children of approximately 2% and 3% respectively, while no statistically significant wage effect is detected for the presence of one child. In Finland and Denmark, wage rates appear to be virtually unaffected by children (Kellokumpu, 2007; Kleven, Landais & Søgaard, 2019). For Norway, Kunze (2020) also concludes that the impact of children on earnings is not statistically significant. The fatherhood wage premium has so far been found to be the largest in North America (Mari, 2019). Lundberg & Rose (2000) report an adjusted wage premium of 9% per child among married fathers in the U.S., while Budig & Hodges (2010) find a wage premium of 6% per child after including an additional control of marital status. However, Glauber (2008) find statistically significant wage premiums to be close to only 1%. For Canada, Cooke & Fuller (2018) show that men experience a wage premium of close to 4% per child.
13 further examined. Yukawa (2015), who provides the only evidence of the wage impact of children on male workers in Japan, shows that fathers on average receive a wage premium of 2.3% per child. The wage increases are statistically significant for one, two and three children of 2.8%, 5.0% and 6.8% respectively, while no statistically significant estimate is found for fathers with four or more children.
2.3 Heterogeneity of Pay Gaps
It is commonly considered that the size of a family wage gap may differ for different groups of parents. For example, a pay gap could depend on the level of education or vary by marital status as well as by job sector (Cukrowska-Torzewska & Matysiak, 2020). To account for such heterogeneity, various subsamples can be used to identify different sources of the wage gaps. In this discussion on how pay gaps can vary by different characteristics of parents, particular focus is paid to whether a motherhood wage penalty varies by educational attainment and by age groups, as this is the main focus for the further investigation of the motherhood pay gap. The heterogeneity of pay gaps from the existing literature from Japan is also reviewed.
2.3.1 The Family Pay Gap by Educational Attainment
14 lower education may experience a higher child penalty. The impact of education on the motherhood wage penalty is not definitive in terms of empirical evidence because it differs between studies and countries. In a meta-analysis by Cukrowska-Torzewska & Matysiak (2020), the motherhood pay gap is found to be greater among less educated women. Todd (2001) also shows that a high level of education greatly reduces the wage gap between mothers and childless women in Canada and the U.S., and it is completely eliminated for highly educated women in Germany. By contrast, Waldfogel (1997) has found that the child penalty increases with the level of education in the U.S. Also Batchelder, Ellwood & Wilde (2010) show that mothers with higher education face a greater wage penalty than those with lower education, and Budig, England & Hodges (2016) conclude that women with high skills and high wages face the greatest motherhood wage penalty in the U.S.
The same question has also been posed by some researchers who investigate the family wage gap among men. Killewald (2012) find no differences in the fatherhood wage premium based on educational attainment among married fathers, while Budig & Hodges (2010) show that white and Latino fathers with a college degree have a significantly higher wage premium than fathers with no degree. However, they find no difference for black fathers by educational attainment. Other factors that have been considered when investigating the heterogeneity of pay gaps among men include marital status, race and ethnicity as well as the employment status or earnings of the partner (Grimshaw & Rubery, 2015).
2.3.2 The Family Pay Gap by Age Groups
15 to refocus on their careers with the decline in childrearing as they and their children grow older, which would narrow the family pay gap among women. On the other hand, if the lack of early investment in human capital and discontinuous employment hinder them from achieving higher-paying occupations, mothers could face an increasing pay gap which continues to widen as they reach midlife.
Empirically, Bianchi, Kahn & García-Manglano (2014)find that the child penalties in the U.S. are larger for women in their 30s than in their 20s. For women with two children, the wage penalty peaks in the 30s and with three or more children, the pay gap is largest for women in their 40s. However, the pay gaps appear to narrow by the 50s where only women with three or more children suffer a statistically significant wage penalty. This is also consistent with the simulated earnings gap by Sigle-Rushton & Waldfogel (2007), who show a notable narrowing of the pay gap by the mid-forties. Viitanen (2014) also finds that for the United Kingdom, the wage penalty is long-lasting and significantly higher for women at the age of 33 compared to 23, however the gap narrows at older ages. Meurs & Ponthieux (2015) argue that an attenuated impact of motherhood on wages later in life indicates that the work effort hypothesis holds, as the demands of childrearing are greater with younger children. Also Anderson, Blinder & Krause (2003) and Cukrowska-Torzewska & Matysiak (2020) argue that a declining wage penalty with age is consistent with the work effort hypothesis.
2.3.3 Heterogeneity of Pay Gaps in Japan
16 instead focuses on how the family wage gap differs according to the children’s ages. The results show a gross wage penalty for children between 7-18 years old, but not for any younger age group. He argues that this likely reflects that most Japanese mothers with very young children choose not to work at all. Regarding the fatherhood wage premium, Yukawa (2015) divides the sample into two cohorts to examine if a wage gap varies by age groups. Having children is significantly shown to increase wage rates for men born before 1960 but has no effect on wage rates for men born after 1960. Moreover, Yukawa (2015) investigate how the gender of children impacts the wages of Japanese men and conclude that the child’s gender has no statistically significant effect on men’s wages.
2.4 Work Culture, Family Policy and Gender Equality in Japan
In terms of nominal GDP, Japan is the third largest economy in the world (IMF, 2021). In 2020, Japan’s labor force consisted of approximately 68.6 million workers with a low employment rate of only 3.0% (World Bank 2021a; 2021b). However, with low birth rates and a declining population, Japan is struggling to secure the future workforce. As one of the causes for Japan’s low birth rates are believed to be due to the challenges faced by women to balance work and family life, Japan needs to address the high levels of gender inequality that are evident in many parts of the society and not least in the labor market. As an example, the persistent ‘glass ceiling' in Japan can be seen by the fact that only 14.7% of senior positions are held by women (World Economic Forum, 2021, p. 14). Further, the income of Japanese women is on average 43.7% lower than that of Japanese men (World Economic Forum, 2021, p. 37) and the gender wage gap is the second largest among the OECD nations at 23.5% (OECD, 2021).
17 to housework and childcare. Typically, they spend less than one-third of the 3+ hours daily average for Swedish and German fathers on unpaid work, and the time spent on such activities does not increase if the mother in the household is working (Akiba, Matsui & Tatebe, 2014, p. 28). Kojima (2013, p. 252) finds that the amount of support women expects from their spouse in raising children influences their fertility decisions. Further, it seems that there is a discrepancy between the desired number of children that Japanese women would like to have and the number of children that they actually do have (Kojima, 2013, p. 251).
18 motherhood penalties. In Japan, the word matahara refers to the maternity harassment commonly experienced by women. The unfair treatment can be both physical and mental and may for example include forcing women to leave their employment (Baba et al., 2021). In the recent study by Baba et al. (2021), it is shown that every fourth pregnant employee has experienced maternity harassment from their workplace. If mothers are less accepted in the labor force, they may be more hesitant to rely on external childcare to be able to return to work sooner after giving birth (Cukrowska-Torzewska & Matysiak, 2020).
To increase gender equality in Japan, the concept of “womenomics” is aimed to promote women's economic advancement (Matsui, Suzuki & Tatebe, 2019). An increased female economic empowerment is believed to benefit the entire economy and has been one of the former Japanese prime minister’s top priorities. The policy goals within womenomics for example include increasing the percentage of fathers who take parental leave, raising the female labor participation rates, expanding the childcare capacity and implementing work-style reforms to limit overtime hours (Matsui, Suzuki & Tatebe, 2019). Kawaguchi, Kawata & Torivabe (2021) argue that the rise in family-friendly government policies that stem from womenomics is the most important mechanism behind the increased share of married women in the labor force. Further, the extension of parental leave has been a major shift in Japan's family policy over the last three decades (Kawaguchi, Kawata & Torivabe, 2021). Considering the Japanese childcare support system, Zhou (2015) states that Japan’s childbirth and childcare leave is not inferior to those of other developed countries. For example, employers must provide paid leave for 6 weeks before and 8 weeks after childbirth. Japanese law further grants both working parents 12 months of parental leave. During the leave, 67% of the wage is retained and paid for 180 days, which drops to 50% for the rest of the leave (Zhou, 2015). However, few fathers actually use the provided leave. Although the share of men who take childcare leave has increased in recent years, only 6.2% of men took childcare leave compared to 82.2% of women in 2018 (Gender Equality Bureau, 2020, p. 11). To increase the percentage of fathers who take paternity leave, many Japanese companies are now incorporating targets for paternity leave into their diversity agendas (Matsui, Suzuki & Tatebe, 2019).
The methodology chapter begins with a presentation of the data from the Japan Household Panel Survey used in the study. Thereafter, the research design is presented, including which econometric models are used and how the analyses are performed. Lastly, the creation of the final panel data set used for analysis is reviewed and the variables that are included in the models are motivated and discussed.
3.1 Japan Household Panel Survey
The analysis of this study is based on survey data from the Japan Household Panel Survey (JHPS/KHPS) for the time period 2004-2018, provided by the Panel Data Research Center at Keio University after submitting an approved application. The panel data set contains two separate surveys running in parallel, the Keio Household Panel Survey (KHPS) and the Japan Household Panel Survey (JHPS), which are merged using an integration program provided by the research center.
The longitudinal surveys are implemented annually and are aimed to reflect the population composition of society, covering a broad range of topics such as employment and education status, academic background, individual attributes and household structures (PDRC, n.d.). Initially, the Keio Household Panel Survey, hereafter shortened KHPS, was implemented in 2004 and consisted of 4,000 households and 7,000 individuals nationwide. To compensate for some respondents dropping out of the sample, additional cohorts of approximately 1,400 and 1,000 survey subjects were added in 2007 and 2012 respectively. In 2009, the Panel Data Research Center began implementing the Japan Household Panel Survey, hereafter shortened JHPS, which was a new survey running in parallel with KHPS. JHPS targets 4,000 male and female respondents nationwide and differs slightly from KHPS by for instance adding a greater focus on questions related to health and medical care. Many questions are common to both KHPS and JHPS, which enables the use of greater sample sizes when handling the merged panel data set.
21 stage of sampling, resident registers for the selected survey areas are used as sampling registers and about 10 people from each area are selected as survey subjects. The respondents included in the surveys are men and women aged 20-69 in KHPS and aged 20 and older in JHPS. Each survey is conducted in January every year and the response rates are between 82.7% and 94.2% during the survey period used.
3.2 Research Design
22 the majority of previous studies and includes regressions from fixed effects models as well as pooled OLS models.The fixed effects models take the following form:
𝑙𝑙𝑙𝑙𝑙𝑙(𝑤𝑤𝑤𝑤𝑙𝑙𝑤𝑤)𝑖𝑖𝑖𝑖 = 𝑋𝑋𝑖𝑖𝑖𝑖′𝛽𝛽 + 𝑤𝑤𝑖𝑖 + 𝑢𝑢𝑖𝑖𝑖𝑖 (1)
where 𝑙𝑙𝑙𝑙𝑙𝑙(𝑤𝑤𝑤𝑤𝑙𝑙𝑤𝑤)𝑖𝑖𝑖𝑖 is the dependent variable observed for individual i at time t, 𝑋𝑋𝑖𝑖𝑖𝑖′ is a 1×k
vector of k independent variables and 𝛽𝛽 is a k×1 matrix of parameters. The component 𝑤𝑤𝑖𝑖
captures unobserved time-invariant individual effects (e.g. career aspirations) and 𝑢𝑢𝑖𝑖𝑖𝑖 is the
error term, which is assumed to have a zero conditional mean. Because 𝑤𝑤𝑖𝑖 is unobservable and
would cause problems with endogeneity if correlated with a regressor, a fixed effects regression eliminates 𝑤𝑤𝑖𝑖(as well as observable time-invariant variables) by averaging the equation across
t and subtracting the mean from each variable in the model (Stock & Watson, 2015, pp. 405-406). The transformed model becomes:
𝑙𝑙𝑙𝑙𝑙𝑙(𝑤𝑤𝑤𝑤𝑙𝑙𝑤𝑤)𝑖𝑖𝑖𝑖 − 𝑙𝑙𝑙𝑙𝑙𝑙(𝑤𝑤𝑤𝑤𝑙𝑙𝑤𝑤)𝑖𝑖 = �𝑋𝑋𝑖𝑖𝑖𝑖− 𝑋𝑋𝑖𝑖 �′𝛽𝛽 + (𝑢𝑢𝑖𝑖𝑖𝑖− 𝑢𝑢𝑖𝑖 ) (2)
This transformed model is estimated by ordinary least squares and controls for time-invariant, unobserved heterogeneity. A pooled OLS model can be formulated with a similar form as equation (1), with the difference that it lacks the component of 𝑤𝑤𝑖𝑖 as it cannot capture the
unobserved and stable characteristics. Running a pooled OLS regression would treat the data as one big, pooled cross-section, but in order for this to produce a consistent and unbiased estimator of 𝛽𝛽𝑖𝑖, any unobserved heterogeneity must be uncorrelated with the regressors
(Wooldridge, 2012, p. 460).
23 Since the aim of this thesis is to analyse the impact of children on wages for both genders, the data set is divided by gender in order to distinguish the wage effects on men and women. Separate but identical regressions are therefore run on both groups. Although the model specifications are based on economic theory and previous research, sometimes a small change in a model leads to considerable differences in the estimates. To help identify which model to choose, the Akaike's information criterion and Bayesian information criterion are used. These measures are widely utilized in model selection and reflect how well a model fits the data (Fabozzi et al., 2014, p. 399). Further, it is possible to measure the parenthood status by either using dummy variables for the number of children or a continuous measure. This study implements both methodologies, for which a motivation can be found in the discussion on creating the data set used for analysis in Chapter 3.3.
24 older. In the subanalyses, only the results from the fixed effects models are presented. All analyses are performed using the software program Stata.
3.3 Creation of the Panel Data Set
In this section, the construction of the data set used for the analysis will be discussed. As the process of handling and managing the data provided by the Panel Data Research Center is a relatively time-consuming and occasionally complex process, it is not possible to thoroughly review assumptions and coding specifications made to construct the final data set. However, the main parts of the process are presented and the inclusion of different independent variables is motivated.
Considering how to best define the groups of parents and nonparents, most studies define the parenthood status by whether the respondent has dependent children or not (Grimshaw & Rubery, 2015). This could either be done by simply restricting the age span of the respondents to ages which usually involve childrearing and being responsible for one or many children, and/or by defining the children of respondents as children who are under a certain age and still live at home. By not restricting the data in any way, the sample will include parents whose children are older and have moved out from home which may confuse the comparison with nonparents (Grimshaw & Rubery, 2015). Pal & Waldfogel (2014) argue that including older women among the nonmothers will diminish the differences between the groups and lead to an underestimation of the magnitude of the motherhood pay gap. As the main aim of this study is to estimate the wage impacts of being responsible for dependent children, this study will define the parenthood status by following previous research. Like Pal & Waldfogel (2014; 2016) and Simonsen & Skipper (2012), the definition of parenthood is in this analysis based on the presence of children under the age of 18 living in the household. The data set is further restricted with an upper limit of the age span, including respondents aged 20-55.
25 children. The dummy variables thus consider children under the age of 18 living in the household, where no children will be used as the reference category. In addition, a continuous measure of the number of children will also be used to be able to define the average wage impact per child.
The workforce coverage includes respondents who report to be working at least 5 hours per week and no more than 80 hours per week. Self-employed workers are excluded, which is a rather common restriction used by e.g. Harkness & Waldfogel (2003), Nestić (2007), Simonsen & Skipper (2012) among others. Nestić (2007) argues that not including self-employed workers is due to the difficulty of separating payments that are generated by capital invested in self-employment from the wages from work.
The dependent variable, the natural logarithm of the hourly wage, is constructed by following the procedure described by Nakamura (2019). First, the respondents are asked to report if their wages are paid daily, weekly, monthly or annually. The respondents then specify their wages according to the payment method they have selected in the questionnaire. For each individual, the annual working hours are obtained by multiplying the reported weekly working hours by 52. This information can then be utilized to calculate the annual labor income for each payment method. The annual income is defined as the sum of the annual labor income and any annual bonus received from work. The hourly wage is thereafter calculated by dividing the annual income by the number of working hours. To obtain the real hourly wage, the nominal wage is converted by using the consumer price index as a deflator with 2015 as the base year. Observations with a real hourly wage of less than 400 yen are eliminated as they lie well below the statutory minimum wage. Lastly, the variable is log-transformed as in most of the research (Cukrowska-Torzewska & Matysiak, 2020). Using a log transformation of the dependent variable allows for easy interpretations of the estimated coefficients in the model. A change in an independent variable by one unit is then associated with an estimated 100β�i change in the
hourly wage on average, ceteris paribus, where β�i is the estimated coefficient of the explanatory
variable i (Stock & Watson, 2015, p. 322).
26 accumulated annually from the age of 18. The variable includes experience from part-time work, full-time work, piecework, self-employment and from working as a family employee. Variables for education are created both as continuous in years and with the use of dummy variables divided into two categories. The dummy variables capture whether the respondent’s highest completed education is junior high school or high school, or if the respondent has completed any higher education including junior college (usually 2-year), vocational school, university (usually 4-year) or more.
Tenure measures the seniority at the present workplace, which is a firm-specific experience commonly included in wage regressions. This independent variable is likely important to control for, as many Japanese firms give regular employees wage premiums based on their seniority at the company (Yamaguchi, 2019). Employer tenure is set equal to the reported tenure when the respondent first enters the sample and is incremented by one for each year the individual reports that he or she is still working at the same company or organization as one year ago. The value of tenure changes to one if the respondent made a job switch or was newly employed during the past year, and to zero for respondents indicating that they have become or are still unemployed. Lastly, the variable that measures the number of years out of the labor market from the age of 18 is calculated as (the respondent's age – the number of years not in school before the age of 18 – the number of years working and/or in school). As women tend to stay out of the workforce to a much greater extent than men when they have become parents, the time out of the labor market has been seen to greatly contribute to the motherhood wage penalty (Anderson, Blinder & Krause, 2002).
27 variables are created respectively. All variables are shown in the descriptive statistics found in Appendix A.
Lastly, civil status is an important factor that can affect the size of the family pay gaps. Since civil status and parenthood are strongly correlated, Cukrowska-Torzewska & Matysiak (2020) argue that an estimated motherhood wage gap that is not controlling for civil statusmay also capture the wage effects of marriage. They discuss that on the one hand, married women could experience a large wage gap due to an increased specialization within the household. On the other hand, a single mother may instead face a greater wage penalty because she is likely to receive less help with parenting. This could affect work productivity as well as working less or selecting into jobs with mother-friendly job characteristics. For men, many studies have shown that marriage is associated with higher wages, which is why it is important to separate the wage effect of parenthood from that of marriage also among men. Further, several studies consider whether the respondent lives together with their spouse or partner to account for household specialization, e.g. Kumlin (2007), Staff & Mortimer (2012) and Anderson, Blinder & Krause (2002). In JHPS/KHPS, the respondents are only asked to answer if they live together with a spouse and not a partner with which they are not married. Although it has become more common for couples to cohabit while unmarried in low-fertility societies (Kiernan, 2004), the proportion of children born to unmarried couples in Japan can be seen as negligible (Bumpass, Iwasawa & Raymo, 2009). A dummy variable of civil status which captures whether the respondent is married and lives together with their spouse is therefore created.
4. Analysis and Discussion
This chapter starts by discussing the descriptive statistics of the full samples of men and women. Thereafter, the empirical evidence from the main analyses is presented together with discussions of the results. Lastly, the results from the subsamples of women are presented and analysed.
4.1 Descriptive Statistics
Table A1 and A2 are found in Appendix A and show the summary statistics of the two samples consisting of women or men used for the analysis. In total, the sample of women consists of 1,861 working women with 8,844 observations. The sample of men includes 2,034 working men and consists of 10,662 observations. The statistics are presented for all women, mothers and nonmothers, with the same sectioning made for the sample with male respondents. Note that women and men who are categorized as nonparents may have adult children, as the parenthood status is determined by whether or not the respondent has children under the age of 18 living in the household.
When comparing the characteristics of working mothers and nonmothers, some differences can be recognized. Nonmothers on average earn a wage that is 10.6% higher than those of mothers. 85.9% of mothers are married and live with a spouse, compared to 44.8% of nonmothers. Regarding human capital characteristics, nonmothers have approximately 3 more years of work experience, 1.6 more years of tenure and they study on average 1 year longer than mothers. There are no noteworthy differences in academic degrees within the sample of women, as about 50% of both mothers and nonmothers have completed a higher education after high school. Mothers spend on average 4.3 more years out of the labor market than childless women. The mean years out of the workforce for nonmothers is still quite high, approximately 4.5 years, however nonmothers may also include women who had birth-related leave in the past. The mean age is approximately 42 years for mothers and 41 years for nonmothers.
29 Women overall appear to work part-time to a great extent, however a significantly higher percentage of mothers report having a part-time job compared to women without children. 66.5% of mothers in the sample work part-time, compared to 41.8% of nonmothers. The high percentage of mothers working part-time can as an aside again also be compared to fathers, where only 1.6% report to be working part-time. No major differences can be distinguished when comparing the job characteristics of company size, industry or occupation. Women are overall most likely to be working in companies with less than 100 employees and the most common occupation is to be employed as a clerical worker. 18.8% report to work in the health and social care industry and 19.3% work in the wholesale and retail industry, which are the most common industries to be employed in.
4.2 Main Analysis
In the main analysis, both pooled OLS models and fixed effects models are presented. As previously discussed, pooled OLS models are included mainly to serve as a comparison that provides insights on the impact of unobserved time-invariant heterogeneity and selection into parenthood. Because pooled OLS models cannot control for unmeasured time-invariant characteristics, the results from the fixed effects models are seen as more reliable and will be used for causal inference. A final fixed effects model take the following form:
log(𝑤𝑤𝑤𝑤𝑤𝑤𝑤𝑤)𝑖𝑖𝑖𝑖 = 𝛽𝛽0+ 𝛽𝛽1𝑂𝑂𝑂𝑂𝑤𝑤𝑂𝑂ℎ𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖+ 𝛽𝛽2𝑇𝑇𝑤𝑤𝑇𝑇𝑂𝑂ℎ𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑤𝑤𝑂𝑂𝑖𝑖𝑖𝑖+ 𝛽𝛽3𝑇𝑇ℎ𝑖𝑖𝑤𝑤𝑤𝑤𝑇𝑇𝑖𝑖𝑟𝑟𝑇𝑇𝑖𝑖𝑤𝑤𝑂𝑂ℎ𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑤𝑤𝑂𝑂𝑖𝑖𝑖𝑖 + 𝛽𝛽4𝐴𝐴𝑤𝑤𝑤𝑤𝑖𝑖𝑖𝑖+ 𝛽𝛽5𝐴𝐴𝑤𝑤𝑤𝑤𝑖𝑖𝑖𝑖2 + 𝛽𝛽6𝐶𝐶𝑖𝑖𝐶𝐶𝑖𝑖𝑖𝑖𝐶𝐶𝐶𝐶𝑤𝑤𝐶𝐶𝐶𝐶𝐶𝐶𝑖𝑖𝑖𝑖+ 𝛽𝛽7𝑆𝑆𝑂𝑂ℎ𝑇𝑇𝑇𝑇𝑖𝑖𝑖𝑖𝑂𝑂𝑤𝑤𝑖𝑖𝑖𝑖 + 𝛽𝛽8𝑊𝑊𝑇𝑇𝑖𝑖𝑊𝑊𝑤𝑤𝑥𝑥𝑝𝑝𝑖𝑖𝑖𝑖 + 𝛽𝛽9𝑊𝑊𝑇𝑇𝑖𝑖𝑊𝑊𝑤𝑤𝑥𝑥𝑝𝑝𝑖𝑖𝑖𝑖2 + 𝛽𝛽10𝑇𝑇𝑤𝑤𝑂𝑂𝐶𝐶𝑖𝑖𝑤𝑤𝑖𝑖𝑖𝑖 + 𝛽𝛽11𝑇𝑇𝑤𝑤𝑂𝑂𝐶𝐶𝑖𝑖𝑤𝑤𝑖𝑖𝑖𝑖2 + 𝛽𝛽12𝑌𝑌𝑤𝑤𝑤𝑤𝑖𝑖𝐶𝐶𝑇𝑇𝐶𝐶𝐶𝐶𝑖𝑖𝑖𝑖 + 𝛽𝛽13𝐻𝐻𝑇𝑇𝐶𝐶𝑖𝑖𝐶𝐶𝑤𝑤𝑇𝑇𝑖𝑖𝑊𝑊𝑤𝑤𝑖𝑖𝑖𝑖𝑖𝑖+ 𝛽𝛽14𝐻𝐻𝑇𝑇𝐶𝐶𝑖𝑖𝐶𝐶𝑤𝑤𝑇𝑇𝑖𝑖𝑊𝑊𝑤𝑤𝑖𝑖𝑖𝑖𝑖𝑖2 + 𝛽𝛽15𝑃𝑃𝑤𝑤𝑖𝑖𝐶𝐶𝐶𝐶𝑖𝑖𝑟𝑟𝑤𝑤𝑖𝑖𝑖𝑖 + 𝛽𝛽16(𝐻𝐻𝑇𝑇𝐶𝐶𝑖𝑖𝐶𝐶𝑤𝑤𝑇𝑇𝑖𝑖𝑊𝑊𝑤𝑤𝑖𝑖 × 𝑃𝑃𝑤𝑤𝑖𝑖𝐶𝐶𝐶𝐶𝑖𝑖𝑟𝑟𝑤𝑤)𝑖𝑖𝑖𝑖+ 𝛽𝛽17(𝐻𝐻𝑇𝑇𝐶𝐶𝑖𝑖𝐶𝐶𝑤𝑤𝑇𝑇𝑖𝑖𝑊𝑊𝑤𝑤𝑖𝑖2× 𝑃𝑃𝑤𝑤𝑖𝑖𝐶𝐶𝐶𝐶𝑖𝑖𝑟𝑟𝑤𝑤)𝑖𝑖𝑖𝑖 + � 𝛽𝛽𝑗𝑗𝐹𝐹𝑖𝑖𝑖𝑖𝑟𝑟𝐶𝐶𝑖𝑖𝐹𝐹𝑤𝑤𝑗𝑗𝑖𝑖𝑖𝑖 20 𝑗𝑗=18 + � 𝛽𝛽𝑗𝑗𝑂𝑂𝑂𝑂𝑂𝑂𝐶𝐶𝑝𝑝𝑤𝑤𝐶𝐶𝑖𝑖𝑇𝑇𝑂𝑂𝑗𝑗𝑖𝑖𝑖𝑖 28 𝑗𝑗=21 + � 𝛽𝛽𝑗𝑗𝐼𝐼𝑂𝑂𝑖𝑖𝐶𝐶𝐶𝐶𝐶𝐶𝑖𝑖𝐼𝐼𝑗𝑗𝑖𝑖𝑖𝑖 41 𝑗𝑗=29 + 𝑤𝑤𝑖𝑖+ 𝐶𝐶𝑖𝑖𝑖𝑖
where 𝑖𝑖 indexes individuals and 𝐶𝐶 indexes time periods. 𝛽𝛽0 is the intercept and 𝛽𝛽1, 𝛽𝛽2, and 𝛽𝛽3
measure the impact of children on parents’ wages. The coefficients 𝛽𝛽7-𝛽𝛽12 capture the effects
of different measures of human capital and 𝛽𝛽13-𝛽𝛽41 capture the effects of job characteristics.
Firm size, occupation and industry represent vectors of dummy variables, each excluding a reference category to avoid a dummy variable trap. The component 𝑤𝑤𝑖𝑖 represents unobserved
time-invariant heterogeneity and 𝐶𝐶𝑖𝑖𝑖𝑖 is the error term. Note also that a similar regression is run
for each sample, where a continuous measure of the number of children is used instead of dummy variables. Beyond that, both model specifications are identical.
4.2.1 The Full Sample of Women
31 for women, where control variables are stepwise added to the models to investigate the effect of separate characteristics. Estimates from the regressions of the final models are presented in Appendix C.
Table 1: The Main Analysis of Women.
Pooled OLS models Fixed effects models .
Number One Two Three or Number One Two Three or of children child children more children of children child children more children
controls) -0.056*** -0.096*** -0.096*** -0.186*** -0.046*** -0.042** -0.094*** -0.133***
Age, Age2 -0.081*** -0.130*** -0.156*** -0.253*** -0.030*** -0.036** -0.065** -0.082**
Age, Age2, Civil
status -0.068*** -0.104*** -0.122*** -0.222*** -0.027** -0.030 -0.058** -0.074**
Above + Education
in years -0.043*** -0.070*** -0.070** -0.144*** -0.027** -0.030 -0.058** -0.074**
Above + Work exp,
Work exp2 -0.038*** -0.053** -0.056* -0.130*** -0.026** -0.031* -0.057** -0.074**
Above + Tenure,
Tenure2 -0.026** -0.047** -0.039 -0.084** -0.028** -0.030 -0.063** -0.066*
Above + Years out -0.026** -0.047** -0.039 -0.084** -0.028** -0.031 -0.063** -0.066*
Job characteristics: Above + Hours worked, Hours worked2 -0.034 *** -0.059** -0.052* -0.113*** -0.051*** -0.058*** -0.111*** -0.134*** Above + Part-time -0.016* -0.009 -0.008 -0.066** -0.044*** -0.040** -0.085*** -0.119*** Above + Interaction terms -0.020** -0.021 -0.019 -0.079** -0.045*** -0.056*** -0.095*** -0.123*** Above + Firm-size -0.019** -0.019 -0.018 -0.073** -0.045*** -0.055*** -0.093*** -0.122*** Above + Occupation -0.020** -0.020 -0.025 -0.073** -0.044*** -0.056*** -0.092*** -0.117*** Above + Industry (final models) -0.020 ** (0.008) -0.020 (0.017) -0.027 (0.019) -0.071** (0.029) -0.044*** (0.009) -0.055*** (0.015) -0.093*** (0.021) -0.117*** (0.030) Notes: The estimates are coefficients of the number of children under the age of 18 living in the household. These are presented from separate model specifications using either a continuous measure or dummy variables, where the omitted category is no children. The dependent variable is the natural logarithm of the real hourly wage in constant 2015 yen. Cluster-robust standard errors are presented in parentheses for the final models. *** denotes statistical significance at a 1% significance level, ** at a
5% significance level and * at a 10% significance level.
Pooled OLS Results
32 mothers tend to invest more time and effort into household chores and childcare, and hence focus less on their careers. Adding human capital controls, the child penalty drops when controlling for years of schooling, work experience and tenure, while the time out of the labor market appears to not affect the wage penalty. In total, human capital explains approximately 62% of the wage penalty per child in the cross-section of women, compared to a model controlling only for age and civil status. Interestingly, adding hours worked per week again increases the child penalty. At first thought, this may appear to not make sense as nonmothers report to work more than mothers, however the estimated coefficient of the variable measuring hours worked is negative. This suggests that working more hours actually lowers the hourly wage, which makes the increased child penalty reasonable if nonmothers work more unpaid overtime. As the majority of workers are paid monthly wages, working overtime without getting paid for those extra hours would naturally lower the hourly wage. In Japan, working unpaid overtime is common and is even said to be a part of Japan’s working culture (Thiruchelvam, 2018). Further adding the dummy variable for whether the respondent works part-time decreases the child penalty drastically, implying that part-time work considerably contributes to the motherhood pay gap. Interaction terms of the weekly hours worked and the dummy variable of part-time work are also added to capture the effect that working longer hours may impact full-time workers and part-time workers differently. The estimates show that the negative wage effect of working more hours is smaller for part-time workers, which is reasonable if full-time workers are more likely to work unpaid overtime compared to part-time workers. The controls of firm size, occupation and industry further lower the wage penalty per child, but the largest effect stemming from job characteristics still appears to come from part-time work. In the pooled OLS model, job characteristics can explain approximately 23% of the remaining wage penalty per child compared to the model including controls of human capital. In the final model, a statistically significant estimate of the average wage penalty per child of 2.0% remains. Further, the results show a wage penalty of 2.0% for mothers with one child, 2.7% for two children and 7.1% for three or more children. These estimates are not statistically significant for the presence of less than three children, however as previously discussed, more confidence will be placed in the fixed effects models as the pooled OLS models are likely to contain omitted variable bias due to unobserved and time-invariant heterogeneity.
Fixed Effects Results
33 approximately 18% per child. This suggests that unobserved heterogeneity is likely to have a large influence on the wage gap. Again, the child penalty decreases when adding the dummy variable of civil status which supports that household specialization may negatively affect the wages of mothers in Japan. When adding human capital controls, the child penalty appears to be unaffected which suggests that these factors do not contribute to the wage gap. This is considerably different from the OLS models. Time-invariant unobserved heterogeneity may nonetheless play a big role in the difference of results between the pooled OLS and fixed effects regressions, which does seem to be the case throughout the analysis. However, controlling for weekly hours worked has a similar effect and increases the wage penalty for mothers. Not including this variable clearly underestimates the motherhood wage penalty. Adding the dummy variable of part-time work again has a noticeable effect on the child penalty, which supports that part of the motherhood pay gap can be explained by the fact that mothers work part-time to a greater extent than childless women. The interaction terms further help to distinguish the effect of working longer hours as a part-time or full-time worker, which slightly increases the wage gaps. Adding dummy variables of firm size, occupation and industry only minimally decreases the child penalty and hence does not seem to help explain the presence of a motherhood pay gap. After the inclusion of all control variables, there still exists a statistically significant motherhood wage penalty of 4.4% per child that cannot be explained. The residual wage gap is also worsened by having more children, as mothers with one, two or three or more children face wage penalties of 5.5%, 9.3%, and 11.7% respectively. All estimates from the complete fixed effects models are statistically significant at a 1% significance level. The results of significant wage penalties for motherhood are thus in line with a great deal of previous research, where the pay gaps are also greater for women with several children. To further analyse the motherhood wage penalty and which mothers bear the cost of parenthood, the sample of women is divided into subsamples in Chapter 4.3 to help investigate the phenomenon.
4.2.2 The Full Sample of Men
34 wage impact per child is statistically insignificant and close to nonexistent at -0.6%, suggesting that men’s wages appear to be virtually unaffected by the presence of children. Table 2 shows the main analysis for men, where control variables are stepwise added to the models. Estimates from the regressions of the final models are presented in Appendix C.
Table 2: The Main Analysis of Men.
Pooled OLS models Fixed effects models .
Number One Two Three or Number One Two Three or of children child children more children of children child children more children
controls) 0.083*** 0.245*** 0.233*** 0.210*** 0.021** 0.063*** 0.032 0.077**
Age, Age2 0.060*** 0.159*** 0.176*** 0.164*** 0.009 0.033* 0.015 0.032
Age, Age2, Civil
status 0.017 0.066** 0.077*** 0.061 0.006 0.029 0.010 0.026
Above + Education
in years 0.020** 0.066** 0.080*** 0.065* 0.006 0.028 0.009 0.025
Above + Work exp,
Work exp2 0.019* 0.061** 0.075*** 0.061* 0.006 0.026 0.008 0.025
Above + Tenure,
Tenure2 0.015 0.039 0.043 0.051 0.006 0.026 0.010 0.023
Above + Years out 0.014 0.039 0.043 0.049 0.006 0.026 0.010 0.023
Job characteristics: Above + Hours worked, Hours worked2 0.011 0.024 0.044 * 0.023 -0.012* 0.009 -0.015 -0.037* Above + Part-time 0.016* 0.032 0.056** 0.039 -0.007 0.013 -0.004 -0.024 Above + Interaction terms 0.016* 0.031 0.056** 0.039 -0.007 0.011 -0.005 -0.023 Above + Firm-size 0.017** 0.022 0.048** 0.046 -0.007 0.010 -0.005 -0.025 Above + Occupation 0.016** 0.019 0.048** 0.044 -0.007 0.010 -0.004 -0.023 Above + Industry (final models) 0.016 ** (0.008) 0.018 (0.017) 0.048** (0.019) 0.042 (0.028) -0.006 (0.006) 0.010 (0.010) -0.003 (0.013) -0.022 (0.019) Notes: The estimates are coefficients of the number of children under the age of 18 living in the household. These are presented from separate model specifications using either a continuous measure or dummy variables, where the omitted category is no children. The dependent variable is the natural logarithm of the real hourly wage in constant 2015 yen. Cluster-robust standard errors are presented in parentheses for the final models. *** denotes statistical significance at a 1% significance level, ** at a
5% significance level and * at a 10% significance level.
Pooled OLS Results
35 effect on men’s wages, and as fathers are much more likely to be married, including civil status into the models explains a significant part of the pay gap between fathers and nonfathers. Including variables of human capital further decreases the child premium to the point where there are no statistically significant estimates of a fatherhood wage premium, however adding controls of job characteristics causes some of the estimates to become significant again. Compared to an unadjusted pay gap, controlling for age, civil status, human capital and job characteristics can explain approximately 81% of the fatherhood wage premium per child in a cross-section of men. The residual wage gaps are statistically significant for the continuous measure of children as well as for the presence of two children, but not for that of one or three or more children.
Fixed Effects Results
4.2.3 Summary Discussion of the Main Analysis
The main analysis of women confirms the results of previous research, as women’s wages are significantly and negatively affected by parenthood. The estimated average child penalty of 4.4% faced by Japanese women lies in between the estimates reported by Dumauli (2019a; 2019b) and Takeuchi (2018). The result may however be most comparable to the longitudinal study made by Dumauli (2019a) that includes full-time and part-time workers, in which case the result suggests that the wage penalty may be slightly smaller than that of 6.1%. Similar to the results of Takeuchi (2018) along with the majority of research around the world, the wage penalty is greater for women with several children. This however differs from the results of Dumauli (2019b), who conclude that the wage gap associated with the presence of one child is consistently higher than that of two children in Japan. In contrast to the analysis of women, the wages of men appear to be unaffected by children as neither a positive nor a negative wage effect can be confirmed. This result contradicts the research of Yukawa (2015), who find a positive wage effect of 2.3% per child for Japanese men. The results showing the magnitudes of the family pay gaps from the complete fixed effects regressions are summarized in Figure 4.
Figure 4: The Magnitude of the Japanese Family Pay Gaps. The fixed effects regressions include control variables of age, civil status, human capital variables and job characteristics. *** denotes statistical
significance at a 1% significance level and the lack of asterisks denotes statistical insignificance.
Comparing the pooled OLS models with the fixed effects models, unobserved time-invariant heterogeneity appears to greatly impact the analyses, which further indicates that using longitudinal data is far superior to cross-sectional analyses when investigating the family wage
-14.0% -12.0% -10.0% -8.0% -6.0% -4.0% -2.0% 0.0% 2.0% Women Men W ag e im pact
37 gaps. In the analysis of men, the fixed effect estimates are consistently lower than the OLS estimates. This suggests that there is a positive selection into parenthood, where factors that are associated with a greater likelihood of having children may be the same factors that contribute to a higher pay. This is consistent with the results found in several European countries by Cooke & Fuller (2018) as well as in Norway by Kunze (2020), however it differs from the U.S. research by Budig & Hodges (2010) and Lundberg & Rose (2002). The mechanism of selection into parenthood is less obvious for women as the fixed effects models initially appear to narrow the pay gaps compared to the pooled OLS models. In the final models however, the estimated wage penalties are greater in the fixed effects models, which points to a positive selection into motherhood. This is also consistent with the results of Dumauli (2019a), who find that the adjusted motherhood pay gap is greater when controlling for time-invariant unobserved heterogeneity. This suggests that stable characteristics which are associated with a higher pay, are also associated with entering parenthood.
Further,the statistically significant estimated coefficient of civil status is negative for women and positive for men and causes the estimated pay gaps to narrow. This, together with the fact that Japan has a sharp division of household work, implies that household specialization explains part of the family pay gaps for both women and men. The job arrangements including working hours as well as working part-time moreover notably affect the child penalty for women. Excluding weekly working hours is seen to underestimate the wage penalty, whereas the fact that mothers are most likely to work part-time contributes negatively to a wider pay gap. As mentioned earlier, part-time work has been shown to lower hourly wages. Because Japanese mothers are more likely to work part-time, this factor is not surprisingly contributing to the motherhood pay gap. However, measures of human capital seem to not help in explaining the motherhood pay gap, which is in contrast to several studies including those of Budig & England (2001) and Staff & Mortimer (2012). However, no previous longitudinal study has tried to explain the relative importance of different factors in shaping the motherhood pay gap in Japan, which is why the results cannot be compared within a Japanese context. For men, the inclusion of human capital measures neither contributes to the fatherhood pay gap in the fixed effects models, however this is not expected from a theoretical point of view.
38 discussed in Chapter 2.1, these results suggest that household specialization and selection into mother-friendly job characteristics by part-time work mainly give rise to the wage penalties faced by Japanese women. For men, the fatherhood wage premium can be entirely explained by time-invariant characteristics, age and civil status. Hence, positive selection into fatherhood and household specialization appear to account for the entire pay gap between fathers and their childless counterparts. The results from the analysis of both women and men thus point to the important impacts of traditional gender roles and the clear division of labor within households in forming the Japanese family wage gaps.
4.3 Further Analysis of the Motherhood Wage Penalty
To further investigate and help decompose the motherhood pay gap, the sample of women is divided into subsamples by educational attainment and age group. As mentioned in the methodology chapter, subsample analyses of men show no evidence of any statistically significant fatherhood pay gaps and these results are not presented or discussed in the thesis.
4.3.1 Subsamples by Educational Attainment
From previous literature, the results of how the motherhood pay gap varies by educational attainment differ. Moreover, as no analysis of whether the magnitude of the Japanese wage penalty varies by education has been performed, the full sample is divided into two subgroups to further examine different sources of the child penalties. One group consists of women with a higher education who have completed junior college (usually 2-year), vocational school or university (usually 4-year) or more. The comparison group has completed high school at most and thus has no academic degree from higher education. In total, the group of women with a higher education consists of 670 individuals and the group that has completed high school or less consists of 748 individuals. Some women may change education groups over the course of the survey, however if they do, their wages are expected to be affected accordingly.