Effects of Fatherhood on Male Wage and Labor Supply in Japan

3.1 Description of data

I use JPSC data collected by the Institute for Research on Household Economics. The JPSC survey began in 1993 with a nationally representative sample of 1500 women aged from 24 to 34 years. Information on these women (and their husbands) has been collected continuously, including data covering employment, income, wealth, expenditures, health, marriage, childbearing, and numerous other topics. Moreover, samples of 500 women aged from 24 to 27 years were added in 1997 and 836 women aged from 24 to 29 years were added in 2003. I also use data on husbands who are married to the surveyed women. ^[6][7] Although the JPSC is a survey of women and its main purpose is to investigate the lifestyles of women, the data also contain rich information on household characteristics, including the labor market outcomes of husbands, such as amount of labor supply, wage information on wives, number of children, and children’s gender. Since the first year of the survey (1993) did not collect data on the annual days of work and salaries, I use the corresponding data for the period between 1994 and 2007.

Table 1 presents summary statistics for my sample, comparing male workers born on or before 1960 (early cohort) with those born after 1960 (later cohort). ^[8] The average number of children is 1.7. ^[9] The share of college graduates is larger in the early cohort, which may seem odd given the increased emphasis on education in later years. This may be because the college enrollment ratio of the cohort born in the early 1970s, who are the children of the “baby boomers” in Japan, is low. ^[10][11] However, it does not necessarily imply a decline in education because the share of those who graduated from junior college, technical schools, or high schools is larger for the later cohorts, than for the early cohorts.

3.2 Endogeneity problems and estimation strategy

While my focus is on the relationship between the labor market outcome of married male workers and the number of children, it is likely that childbirth depends on various unobserved factors that are also correlated with labor market outcomes. For example, if unobservable individual characteristics, such as preference, personal traits, or innate ability affect both the labor market outcome and the number of children, the standard ordinary least squares (OLS) estimation cannot clarify the causal effect of childbirth on labor market performance. If the number of births depends on unobservable individual characteristics that also affect income and labor supply, it would be incorrect to interpret the correlation between the number of children born and the labor market outcome as a causal effect of childbirth on labor market performance.

The previous literature deals with this endogeneity problem mainly by two methods. The first approach is to use instrumental variables (IV). If we can find instruments that are correlated with the number of children but not correlated with unobservable individual traits, we use these variables as IV and can gain consistent estimators. For example, Angrist and Evans (1998) used the sex composition of the first and the second child as an instrument for the number of children. While sex composition is randomly determined and thus, is not correlated with labor market outcomes of parents, the sex composition of children has a considerable effect on the number of children, given that it drives the parents’ decision on whether to give birth to the third and any subsequent children. However, their IV can only be used to estimate the impact on the third and subsequent children. Thus, I do not adopt their method, as some of my samples have more than three children.

The second approach is to control the unobserved fixed effects (FE). If unobservable individual traits correlated with the number of children are time invariant, then by adding an individual specific term to the explanatory variables, I can control the effect of time invariant heterogeneity, and get a consistent estimator. Following Korenman and Neumark (1991), Waldfogel (1997), Lundberg and Rose (2002), and Choi et al. (2008), I adopt this approach to deal with the endogeneity problem. Although the FE method is standard in the literature, we should note that we might not remove the potential bias completely, even if we can control the time invariant traits. For example, if workers have children when they expect future wage growth, causality runs from the wage to the number of children. When the impact of expected wage growth on fertility decisions is not negligible, the FE estimator of impact on wages has a positive bias. Unfortunately, we cannot remove the impact of expectation on the wage growth, which is the limitation of the empirical method used here, and is shared by the abovementioned studies. Hence, we must interpret the estimated impact of childbirth on wage with caution. A similar argument can be applied to labor supply analysis, but I believe that the estimated impact on hours of work has little bias, because the hours of work are relatively stable in Japan.

5.1 The effect of children on wage

First, I examine the impact of fatherhood on wage. Table 2 presents the estimated effect of childbirth on wage in the total sample. Columns (1)–(4) report OLS estimators, and columns (5)–(8), FE estimators. I provide the results of the linear and nonlinear models. Columns (3), (4), (7), and (8) control the effect of industry, occupation, firm size, and age, respectively. When I control for the unobserved FE, the estimated impact of additional children on wages is approximately 2.3% (see columns 5 and 7). ^[13] Since the FE coefficients are larger than the ones obtained from OLS, a negative bias in the OLS estimators is likely. One interpretation is that the number of children and innate earning ability are negatively correlated. Japanese male workers who earn higher incomes tend to prefer having fewer children. However, since the difference between the OLS and FE estimators is not large, I believe that the correlation between the number of children and unobserved earning ability is not large.

The estimated results of nonlinear models indicate that having the second or the third child also increases wage at a 1% significant level. The additional impact of the birth of the second child is approximately 2.2% and that of the third child is in the range 2–1.8% when controlled for unobserved heterogeneity.

The finding that childbirth has significant impact on the wage of fathers is somewhat different from the result of Kawaguchi (2005a), who finds that only the birth of the third child has a significant impact on the wage of fathers. This disparity probably comes from a difference in sample periods, since I have similar results to Kawaguchi (2005a), if I only use the sample between 1991 and 2000. I divided the total sample into two cohorts – those born in or before 1960 and those born after 1960 – to examine whether the effects of children on wages differ among cohorts. As argued in Section 1, these cohorts seem to face entirely different labor market environments that may cause differences in the impact of childbirth among cohorts. Table 3 presents the estimated impact by cohort, which controls the effect of industry, occupation, firm size, and age, respectively. Columns (3) and (4) for FE estimations indicate a greater impact of childbirth on wage than the estimations of the full sample. The birth of children increases the wage rate of male workers by approximately 4% on average and is significant at the 10% level. Although the estimated impact in the early cohort is larger, the estimation results are less precise than those for the full sample, partly because of the decrease in sample size. Column (4) indicates that having the first and second children increases the fathers’ wage by approximately 6.1%, but these impacts are not statistically significant. The birth of the third child alone has a significant positive effect on wage. Since there is a large difference between the two estimators, the negative bias of the OLS estimator is larger in the early cohort. This may indicate a clear tendency for poor workers in the early cohort to have many children, which may cause the negative coefficient of the number of children in the OLS estimation.

Columns (5)–(8) in Table 3 present the estimated impact in the later cohort. When I control for unobserved heterogeneity, the estimated impact of childbirth on wage is less than 1% and not significant. In the later cohort, the OLS estimators predict a larger impact than the FE estimators. Therefore, in the later cohort, OLS estimators have a positive bias; this implies that poorer workers in the later cohort tend to have fewer children, which is contrary to the corresponding observation in the early cohort. ^[14]

From these results, I confirm that the birth of children has significantly positive effects on fathers’ wage in Japan. The wage premium of fatherhood is 2.3% on average and nearly linear for the number of children. However, the estimated impacts differ among cohorts.

While the wage premium of fatherhood is substantial in the early cohort, a positive effect of childbirth on wage is not observed in the later cohort. These results indicate that the wage premium of fatherhood has in fact been decreasing in recent years. It would reflect that the long-term impact of childbirth is greater than the short-term impact, because a large part of an increase in wage is realized only after several years after childbirth by promotion later on in their careers. Otherwise, the recent trend in Japanese labor market would affect the work environment differently across cohorts. Sasajima (2009) points out a decrease in family benefits after the long-term stagnation during the 90s. If the younger workers have a lesser chance to get a job in a firm that provides generous family benefits than the elder workers, which is often argued, and arises due to labor market rigidity, the difference in impact of childbirth on wage comes from the difference in family benefits provided by the firm. I also find that biases of the OLS estimator due to unobservable characteristics are substantial and differ among cohorts. In the early cohort, those who have more children are likely to earn lower wages, but in the later cohort, rich workers tend to have many children.

5.2 The effects of children on labor supply

Next, I examine how childbirth affects the labor supply of male workers. Table 4 presents the estimated impact on the labor supply in the total sample, where both estimators predict that childbirth increases the annual hours of work significantly. The estimated impact of childbirth on annual working hours is 69 hours on average, if I control for individual FE. The additional impact of the birth of the second child also increases the annual hours of work by 92–101 hours, a value that is not very different to the estimated impact of the birth of the first child (73–77 hours). While the birth of a child generally increases the labor supply of fathers, the birth of the third child does not do so. Since the FE coefficients are larger than those of the OLS estimator, the latter may have a negative bias.

Table 5 presents the estimated impact of childbirth on the labor supply by cohort group. It is clear that the estimated impacts on the early cohort are very different between the OLS and FE estimators. The FE estimation in Columns (3), and (4) shows that the childbirth would have almost none or a negative effect on the annual hours of work. The observed large positive biases of OLS estimators imply that male workers who tend to work longer are likely to have many children. Considering this result with that of Table 3, I observe the tendency that workers in the early cohort who earn lower wages and work longer tend to have more children. The nonlinear model shows that the birth of the second child reduces working hours significantly. Columns (5)–(8) in Table 5 present the estimated effects on labor supply in the later cohort. Contrary to the results in the early cohort sample, the birth of children increases annual hours of work for men in both the OLS and FE models. The average impact of childbirth on annual work hours is 89 hours, if I control the effects of unobserved characteristics. OLS estimators would have a negative bias in the later cohort sample, which is contrary to their positive bias in the early cohort sample. In the later cohort, male workers who provide large amounts of labor supply are likely to have fewer children. The estimation of nonlinear models clarifies that the impact of childbirth is nonlinear, and that the impact of the birth of the third child is negative and not significant.

These results indicate that although childbirth significantly increases men’s annual hours of work in Japan, its impact differs among cohorts. When unobserved heterogeneity is controlled, childbirth has no significant effect on the annual hours of work in the early cohort, but has a significantly positive impact on the later cohort. Therefore, contrary to the impact on wages, the effect of childbirth on the annual hours of work has been increasing in recent years. Moreover, the direction of bias of the OLS estimator differs between the two cohorts, which would suggest that the relation between unobserved ability or preference, and the fertility decision, has changed across cohorts. In the early cohort, those who work for a longer time, but earn low wages, are likely to have more children, but the opposite is true for the later cohort.

Now, I can summarize the impact of childbirth on labor market outcomes as follows. First, while the birth of children significantly increases the wage rates of male workers in the early cohort, it has no significant impact on their labor supply. Second, the birth of children increases the annual hours of work in the later cohort, but the wage premium of fatherhood cannot be observed in the later cohort. Third, in the early cohort, there is a strong tendency for those who earn lower wages and work longer hours to have more children. On the other hand, in the later cohort, workers who earn higher wages but work fewer hours tend to have more children. These results clearly show the effects of childbirth on both the wages, and the labor supply, and that the tendencies of behavior towards childbirth are very different between cohorts. Interestingly, the estimated increase in hours of work is about 3.3% of average annual hours worked. ^[15] This number is close to the estimated premium of childbirth on the wage rate in the early cohort. Hence, the estimated increase in the annual earnings due to childbirth is similar across cohorts. One interpretation is that male workers in the later cohort find that it is difficult to earn higher wages when they have a child, because they face tight labor market conditions, and so, they increase working hours in order to bear the costs of childcare.

5.3 Discussion

My empirical findings suggest that childbirth increases the wage rates of male workers by approximately 2.3%, and annual working hours by approximately 69 hours in Japan; moreover, the impact of childbirth differs among cohorts. In this section, I briefly discuss the cause of these impacts. Since the intensity effect predicts a negative impact of childbirth on both the labor supply and wage, these results imply that the intensity effect is not strong in Japan. Then, can the observed positive impact of childbirth be interpreted as a result of the specialization effect? The specialization hypothesis predicts that childbirth positively affects the labor market outcomes of male workers, through sexual division of labor in the household.

In order to test the validity of the specialization hypothesis, I examine whether the estimated impact of childbirth changes, when I control for the effect of the wife’s employment status. Lundberg and Rose (2000) showed that female work status affects male labor market outcomes. Here, I assume that the effects of the wives’ work status on husbands’ labor market outcomes capture the specialization effect, at least partially. If the positive impact of childbirth on the labor market outcome of husbands is caused largely by specialization in the household, which involves changes in the work status of wives, the estimated impact of childbirth becomes substantially small, if I control the variation in the work status of wives. I test this hypothesis in the following manner. However, note that the employment status of wives cannot capture the degree of specialization in the household, since specialization does not necessarily imply the exit of wives from the labor market. I can control for only a part of the specialization effect by including the employment status of wives in the empirical models.

I estimate the following extended linear model

where W_it indicates the wife’s employment status in period t, which takes value of one if the wife works, and zero otherwise. Table 9 presents the results of childbirth effects on husbands’ wages, when I control for the wives’ employment status. It shows that the labor market participation of wives decreases the wage rate of husbands by approximately 1.7%, which is consistent with the specialization hypothesis. The impact of childbirth decreases slightly to 2%, but remains significantly positive even if I control for the wives’ employment status. Table 10 presents the results of childbirth effects on the male labor supply when I control for the wives’ employment status. It shows that the estimated impact of childbirth does not change, even if I control for this element. Therefore, the changes in the wives’ employment status after childbirth does not seem to be an important source of estimated impact of childbirth. These results imply that the income effect, which predicts that childbirth increases labor supply or work effort because the value of income increases after childbirth, is the dominant factor that drives the premium of fatherhood. However, one cannot reject the specialization hypothesis with these results, since the wives’ employment status is an incomplete measure of specialization. Wives would cause a decrease in the labor supply even if they do not quit their jobs, or if husbands would do less housework after childbirth. Without more information on household behavior, the importance of specialization within a household cannot be evaluated precisely.

It is also important to identify the source of the observed change in wage or hours of work after childbirth, especially whether the estimated impact of childbirth can be explained by the job turnover of fathers. If I can control the fixed effects of the firm, the importance of turnover could be estimated by comparing the baseline coefficient of number of children with those where I control the firm fixed effects. The difference between these two estimators captures the effect of childbirth when changing firms, that is, the impact of job turnover. However, I do not have data to identify the firm where the worker is currently employed. Therefore, I simply compare the baseline result with those where I restrict the sample to workers who did not change firms during the sample period. Table 11 shows that the estimated impact of childbirth on wages among these workers is not significant, and the impact on hours worked is less than the result in the full sample. Although this result seems to imply that childbirth affects labor market outcomes of their fathers mainly through the job turnover of fathers, I cannot accept the result at face value, since it does not rule out the selection bias problem. However, it could be said that the effect of turnover is not negligible in explaining the effect of childbirth on the labor market outcomes of their fathers. ^[16]

I also confirm whether changes in the type of jobs, that is, regular employment or non-regular employment, is important for changes in the labor market outcomes of fathers after childbirth. As shown in Table 12, the estimated impact does not change even if I control for the type of employment. Therefore, the change in type of employment contract would not alter the impact of childbirth on the labor market outcomes of their fathers.

Table 13 shows the result when I control for savings and the value of the portfolio, which are proxies for wealth of the household, to examine whether the income effect explains the increase in hours of work due to childbirth. The table shows that the estimated impact does not change if I include these variables as control variables. Therefore, the income effect would not play a major part in the relationship between hours of work and childbirth.

Finally, I investigate the possibility that the estimated impact of childbirth on wages captures the effect of family allowance pay only. Although there is only limited information on family allowance, I believe that the family allowance pay does not explain the entire impact. According to the survey on wages (chingin-zijoutou sougou chousa) conducted by the Ministry of Health, Labor and Welfare, approximately 80% of large firms – which have over 1,000 employees – pay family allowance to workers. The average amount of child allowance provided by these firms is approximately 7,000 yen, which is roughly equal to 2% of the average monthly wages in our sample. On the other hand, the survey on wage and retirement allowance of small firms in the Tokyo metropolitan area (chushou-kigyou no tingin taishokukin zijou), conducted by the Tokyo Metropolitan Government Bureau of Industrial and Labor Affairs, shows that only 65% of small firms (with less than 300 employees) pay family allowance, and the average amount of child benefit is approximately 5,000 yen. Since the estimated impact of childbirth in this study captures the average effect of all samples, and about 40% of our sample is employed by companies that have less than 100 employees, family allowance pay can explain only a part of the estimated impact of childbirth on wage. ^[17] Moreover, the family allowance pay effect does not explain why the impact of childbirth on wage differs across cohorts.

I now compare my results with those of Lundberg and Rose (2002) and Choi et al. (2008) to ascertain differences and similarities in the impact of childbirth on male wages and their annual work hours among Japan, the US, and Germany. Figure 3(a) shows that the impact of childbirth on wages is larger in the US and Germany than in Japan. Figure 3(b) shows that the impact of childbirth on labor supply is larger in Japan than in the US and Germany, except for the birth of the first child in the US. Can these differences in the impact of childbirth among Japan, the US, and Germany be interpreted as being caused by differences in labor market institutions and customs? Burgess (1999) showed that the proportion of employed males in the age group of 26–45 years in temporary jobs (1 year or less), was 20% in the US, 16.34% in Germany, and 4.32% in Japan. Further, Sousa-Poza and Henneberger (2004) showed that the turnover intention rates in the US and Germany are larger than that in Japan. Since the labor market in the US and Germany is more flexible compared to Japan, workers in these countries may be able to change jobs when they have children. On the other hand, the relationship between wages and performance is relatively weak in Japan, which may account for the differences in the impact of childbirth on wages among the US, Japan, and Germany. However, since the impact of childbirth on wages in the early cohort in this study is almost the same as that in the US, the gap in the wage premium of fatherhood among the three countries may be caused by the long-term economic stagnation during, and after the 1990s in Japan; thus, it cannot be explained conclusively by institutional factors.

Figure 3:

Effect of children on (log real hourly) wage and annual hours of work: A comparison among Japan, Germany, and the U.S. (a) Log real hourly wage. (b) Annual hours of work.

Figure 4:

Distribution of number of sons and daughters. (a) Born before 1960, (b) Born after 1960.

In order to ascertain the robustness of my findings from the cohort analysis, I test whether my results change if the sample is segmented differently. I divide the total sample into three cohorts (born in 1960 or earlier, born between 1961 and 1965, and those born after 1965) and examine whether the effects of childbirth on male wage and labor supply show a similar trend to the one seen in earlier sections. Table 14 shows the effect of childbirth on male wages and labor supply for the three cohorts. Childbirth is significant at the 5% level and has positive effects on male wages for the earliest cohort alone (born in 1960 or earlier). For this cohort, childbirth leads to an increase in male wages by approximately 3.9%. In the second cohort (born between 1961 and 1965), childbirth still has a positive but smaller effect on wages of males. Further, in the latest cohort, childbirth is not significant and has a negative effect on male wages. These results indicate that the wage premium has been decreasing in recent years, and this trend is observed in the results of two cohorts (born in 1960 or earlier and born after 1960). On the contrary, childbirth is significant for the latest cohort and has a positive effect on the labor supply. Childbirth increases annual work hours by approximately 108 hours in the latest cohort. In the cohort born between 1961 and 1965, childbirth also has a positive effect on the labor supply, but the estimated impact of childbirth on labor supply is small. Moreover, childbirth has a negative and insignificant effect on the labor supply in the earliest cohort. These findings are consistent with my main results. Therefore, it can be stated that the above findings do not depend on the definition of cohort groups.

5.4 The effects of child gender on wage and labor supply

Finally, I examine whether differences in the gender of the child affect the impact of childbirth on labor market outcomes. Table 6 presents the effects of child gender on wages. It shows all results that correspond to specifications (i)‐(iv). Table 6 shows that, irrespective of model specifications, the impact on wages is not significantly larger for either gender. Child gender has no significant effects on men’s wages in either the full sample or both the cohort samples. These results have also been observed in the US; Lundberg and Rose (2002) showed that difference in child gender had no significant effect on men’s wages.

Table 7 shows the results pertaining to the effects of child gender on the male labor supply. The effect of number of sons is larger than that exerted by the number of daughters, but the difference is not statistically significant. This result does not change even if we divide the sample into the early cohort and the later cohort. Therefore, children have no significant effect on men’s annual hours of work in the early cohort sample, based on the gender of the child.

According to these results, child gender does not seem to have any significant effects on men’s labor market outcomes in Japan. This result is contrary to that of Lundberg and Rose (2002) for the US, which showed that the birth of sons has a significantly larger impact than the birth of daughters. This finding might seem somewhat surprising, because a strong preference for sons is observed in many Asian countries (e.g. Das 1984; Arnold and Liu 1986; Larsen et al. 1998). However, Kureishi and Wakabayashi (2011) find that son preference is not observed among the cohort born after 1940 in Japan. One possible explanation of this is that the generous social security system makes economic ties between elderly parents and their adult children weak, although cultural ties are strong, as in other Asian countries. ^[18] The difference in the generosity of social security may be able to explain the difference in the gender effect between Japan and the US, if the gender effect of childbirth comes from old-age support motive of parents, which is one possible explanation of the gender effect, Lundberg and Rose (2002) provide. If sons are less likely to take care of their parents than daughters, fathers who have sons would work harder than those who have daughters to provide for them in their old age. In Japan, parents would have less old-age support motive than those in the US because social security guarantees the living expenses of parents in their old age.