2.1 Relationship between women's income and childbirth

The hypothesis that emphasizes the substitution effect of female income on fertility presumes that the burden of childrearing is mostly allocated to women.Footnote ³ Earlier, Willis (Reference Willis1973), proposing a static economic theory of lifetime marital fertility, further reports that a wife's education is negatively related to the number of children. Butz and Ward (Reference Butz and Ward1979) claim that the mother's time is raw material for childcare services; thus, an increase in female wages hinders childbirth. Similarly, Heckman and Walker (Reference Heckman and Walker1990) suggest that a higher income for a woman delays the timing of childbirth, whereas a higher income for a man speeds up childbirth. Recently, Santarelli (Reference Santarelli2011) examines data from Italy, which, similar to South Korea, is a very low fertility country, and further finds that the employment of women greatly reduces the possibility of childbirth.

However, traditional premise is somewhat different from the recent situation. Although childbirth is biologically exclusive to women, male participation in postnatal care is expanding in the most developed countries. More importantly, there have been many developments in institutional devices that help women reconcile their work and family, such as maternity leave, childcare subsidy, and flexible work time, which may have greatly reduced the substitution effect from economic losses before and after childbirth. Under these circumstances, in the 2000s, an argument was proposed that the impact of women's income on childbirth is not negative. Andersson (Reference Andersson2000) demonstrates that fertility is pro-cyclical in Sweden. Similarly, Sobotka et al. (Reference Sobotka, Skirbekk and Philipov2011) assert that the reason for fertility being pro-cyclical is that women perceive childbearing to be a risky strategy during recession. Kornstad and Rønsen (Reference Kornstad and Rønsen2018) suggest that the possibility of childbirth decreases with income at first; however, if income exceeds a certain level, the possibility of childbirth increases in Norway (U-shaped relationship).

Notably, many studies argue that the effect of female income on childbirth depends on social systems because they determine the size of opportunity cost of childbearing. Baizan (Reference Baizan2004) argues that the employment of women in Italy, Spain, and Britain is negative for childbirth, but it is independent of fertility in Denmark. Subsequently, Andersson et al. (Reference Andersson, Kreyenfeld and Mika2009) report that in Denmark, which has a welfare state system, earning a sufficient income has become a necessary condition for women's childbearing, but in Germany, where the role of a female homemaker is emphasized, women's income and childbirth have a negative relationship. Köppen (Reference Köppen2006) investigates Germany and France that have markedly different work–family balance policies. In Germany, the positive effect of a wife's educational level on childbirth is greatly weakened when controlling her husband's educational level. In France, by contrast, women have less difficulty performing both work and childcare simultaneously compared to German women, which led to the greater income effect of education on childbirth.

In a similar context, a growing body of quasi-experimental literature has investigated the change in fertility resulting from major welfare reforms. Some of them focus on the heterogeneous effects of policy change on childbirth. For instance, in 1990, Austria increased the duration of the maternity leave from 1 year to 2 years. However, the positive effect of the reform was stronger for low-wage women than it was for high-wage women as flat rate cash transfer implies a higher earnings replacement ratio the for the lower-income group [Lalaive and Zweinmuller (Reference Lalaive and Zweinmuller2009)]. On the contrary, Germany introduced a new paid maternity leave such that its benefits continuously increase in earnings in 2007. As the reform leads to greater benefits to higher-income women, the reform's positive effect on fertility is driven by women at the upper end of the earnings distribution [Raute (Reference Raute2019)]. According to Cygan-Rehm (Reference Cygan-Rehm2016), the lowest-income mothers experience a loss in the total benefits because of a shorter entitlement period. Consequently, the new system of Germany initially leads to a postponement of childbirths, with the negative effect further persisting, particularly for the lowest income women. However, a cross-country analysis performed by Olivetti and Petrongolo (Reference Olivetti and Petrongolo2017) on 30 OECD countries reports that the effect of parental leave on fertility is quantitatively negligible.

In terms of studies conducted on South Korea, Kim Choe and Retherford (Reference Kim Choe and Retherford2009) suggest that the rapidly rising levels of women's education contributed to decline in fertility. Ma (Reference Ma2013) proposes that women's withdrawal from the labor market is a sign of family expansion. However, the study also finds that there exists the possibility of a positive relationship between earnings and fertility because the fertility rate of private sector employees appears to be sensitive to business cycles. In a later paper, Ma (Reference Ma2016) illustrates that the effects of women's educational attainment on fertility lessened when the husband's educational attainment was considered. This finding is in line with Köppen's analysis (2006) on Germany, where the husband's economic power plays an essential role in family expansion. Notably, there are multiple studies that investigate the implications of high education cost on the relationship between income and childbirth. According to Anderson and Kohler (Reference Anderson and Kohler2013), child education is extremely expensive in South Korea, such that the average couple cannot afford to raise more than one or two children. The trend of high parental investment in childhood education at present is assumed to exemplify the notion of quality over quantity. Additionally, Kim et al. (Reference Kim, Tertilt and Yum2019) propose status externalities as a new reason for South Korea's low fertility. Because parents care about the education of children relative to other parents' children, they tend to over-invest in education (quality) and under-invest in fertility (quantity). These studies may support the positive correlation of (family) income and childbirth but do not relate mothers' income with their childbearing under the government's policy change toward a more family-friendly society.

2.2 South Korea's first Basic Plan for Low Fertility and Aged Society

South Korea's work–family balance system began to strengthen in earnest in the mid-2000s. In particular, when the birth rate reached a record low in 2006, the government promulgated the first Basic Plan for Low Fertility and Aged Society (BPLA) and continued to implement policies to promote fertility [Frejka et al. (Reference Frejka, Jones and Sardon2010)]. The plan expanded subsidies for the costs of childrearing and education, day care for children, and after-school programs. Taxes and health insurance costs were lowered for families with young children. Additionally, maternity leaveFootnote ⁴ and childcare leaveFootnote ⁵ were significantly expanded. Childcare leave that had been previously applied only to parents of children under the age of 1 year was extended to those with children up to 3 years of age. Leave benefits were also substantially increased. Because those devices facilitate a stay in or return to the labor market before and after childbirth, the better the policy is applied the less lifetime income loss results from childbearing.

If the work–family balance system is successful in mitigating the substitution effect of women's income, then the relationship between women's economic power and childbirth would be likely to change [Feyrer et al. (Reference Feyrer, Sacerdote and Stern2008)]. Lee (Reference Lee2009) notes that the number of childbirths increased 2 years after the government set up the first BPLA in 2006. However, he argues that to conclude that the fertility rate has reached its lowest point is premature. The BPLA made 2006 a turning point in the history of South Korea's maternity protection system although the change in system was not sudden but gradual as shown in Table 1 and Figure 1. Therefore, we assume that there has been a change in the relation of women's income and their childbearing since the year 2006.

Figure 1.

Change in the number of women on maternity leave. Source: Ministry of Employment and Labor [Korea Employment Information Service (2017)].

Table 1.

Development of maternity and childcare leave in South Korea

^a Entitled establishments are businesses with the number of employees under 500 (manufacturing), 300 (i.e., construction, transportation, business services, and medical services), 200 (i.e., retail and wholesale, accommodation, food services, finance, and arts and sports), or 100 (other sectors). For convenience, the exchange rate between the Korean won and US dollar is set to 1,000 won = 1 dollar.

Notably, the different benefit structures of maternity protection system may lead to the policy exerting a heterogeneous impact on fertility according to women's income. For instance, maternity leave is expected to have a greater impact on high-income women than low-income women because the benefit covers the entire salary. By contrast, childcare leave with a fixed amount of benefit (before 2011) can exert a greater impact on low-income women than high-income women if both are eligible. However, childcare leave is not available for most low-income women in South Korea as presented in section 4.1. Thus, the overall impact of the policy change, which increased childcare leave benefit in 2006, would be greater for high-income women. Conversely, childcare benefit was changed from a fixed amount to a proportion of the mother's earnings in 2011. This reform would yield increased benefits for higher-income women, which is similar to Germany's introduction of a new maternity leave scheme in 2007.

3.1 Survival analysis

In the basic form of the survival analysis model, the hazard is defined as the conditional probability that an event will occur soon after the event has not occurred. The probability that an event does not occur until a certain point in time is called survival function [S(t)], which is used to calculate the hazard function [h(t)] of the following formula:

(1)$$\eqalign{{\rm \; }h\left( t \right) = & \; \mathop {\lim }\limits_{\Delta t\to 0} \displaystyle{{\Pr \left( {t \le T \le t + \Delta t{\rm |}T > t} \right)} \over {\Delta t}}\; \; \left( {T = {\rm duration\; time}} \right) \cr = & \ \displaystyle{{\,f\left( t \right)} \over {S\left( t \right)}} = \; \displaystyle{{\,f\left( t \right)} \over {1-F\left( t \right)}}\; \quad \; \left( {S\left( t \right) = \Pr \left( {T > t} \right),\; f\left( t \right) = {F}^{\prime}\left( t \right)} \right)} $$

This study uses the Cox proportional hazards model, a representative form of a semiparametric model, to analyze the hazard of childbirth. The underlying assumption of the model is as follows:

(2)$$h_i( t ) = h_0( t ) {\rm exp}\mathop \sum \limits_{k = 1}^S \beta _kx_{ik}$$

Here, h _i(t) refers to the hazard of an event occurring for an individual i, and h ₀(t) is a baseline hazard function that is a function of time t. S is the number of explanatory variables that affect the hazard. In this case, the individual i and the other person j have different values for the explanatory variables; furthermore, the difference can be represented as a ratio of the hazards of the two individuals:

(3)$$\;\displaystyle{{h_i( t ) } \over {h_j( t ) }} = \displaystyle{{h_0( t ) {\rm exp}\mathop \sum \nolimits_{k = 1}^S \beta _kx_{ik}} \over {h_0( t ) {\rm exp}\mathop \sum \nolimits_{k = 1}^S \beta _kx_{\,jk}}} = {\rm \;exp}\mathop \sum \limits_{k = 1}^S \beta _k( x_{ik}-x_{\,jk}) $$

In equation (3), h ₀(t) disappears when calculating the hazard ratio; thus, estimation is possible for the influence of the explanatory variable on the hazard without any additional assumptions on the baseline hazard function. Assuming a single explanatory variable, we express individual i's likelihood as formula (4); thus, partial likelihood is measured by multiplying the likelihood value of each individual and appears as formula (5). Coefficient (β) can therefore be estimated using maximum likelihood estimation:

(4)$$\;L_i = \left[{\displaystyle{{\exp ( {\beta x_i} ) } \over {\exp ( {\beta x_i} ) + \exp ( {\beta x_{i + 1}} ) + \cdots + \exp ( {\beta x_n} ) }}} \right]^{\delta _i}$$

where  δ _i = 1  if  event  occured, δ _i = 0 if  censored.

(5)$${\rm P}{\rm L}_i = \; \mathop \prod \limits_{i = 1}^n \left[ {\displaystyle{{\exp \left( {\beta x_i} \right)} \over {\mathop \sum \nolimits_{\,j = 1}^n M_{ij}\exp \left( {\beta x_j} \right)}}} \right]^{\delta _i}$$

where M _ij = 1  if  t _j ≥ t _i, M _ij = 0 if  t _j < t _i.

When the timing of childbirth is analyzed, it is necessary to set the starting point for women to be able to give birth [Newman and McCulloch (Reference Newman and McCulloch1984)]. Research on European or United States' cases have usually set the starting point at the age when a female can become pregnant (e.g., aged 15 years), rather than the time of marriage, because in the West, marriage and pregnancy are not necessarily related. In South Korea, however, where births outside of marriage account for only 1–2% of the total births, an appropriate starting point should be the time of marriage.Footnote ⁶

3.2 Data

The data used to analyze the relationship between women's income and childbirth are from the KLIPS (Korea Labor and Income Panel Study) produced by the Korea Labor Institute in South Korea. A detailed database of employment and labor conditions and individual characteristics, such as the age, gender, residential area, income, and educational level, is available. In this study, data accumulated over several periods from marriage to childbirth are essential; thus, the KLIPS data are the most suitable. The samples are limited to married women aged 20 years or older and below 45 years. The Population Census conducted by the Korean Statistical Office in 2015 states that nearly 90% of South Korean women bear either one or two children. Therefore, the analysis is limited to the first and second childbirths. The basic statistics of the sample of 1,137 women to analyze their first childbirths are presented in Table 2.

Table 2.

Basic information of the sample of women

The average age at marriage in South Korea is much higher than that in other countries. This tendency is mainly a result of women's high levels of education, men's mandatory military services for 2 years, and increased competition for decent jobs suitable for young people. Statistics are based on the information in the year when women were surveyed for the first time, except for age at the time of the birth of the first child.

However, a caveat when using women's income as the explanatory variable is that many women are unemployed. We cannot measure the actual income of unemployed women. As a result, some studies have used predicted wages calculated by separately estimated wage equations mostly using age and education information. However, although age and education have a marked effect on income, many other critical determinants of income are missing. To clarify, in South Korea where nearly 80% of young women graduate from college or university, it is difficult to assert that all these women have the same economic power. Therefore, educational level is less helpful in predicting economic potential. Additionally, years of education are not informative in predicting one's income because it is rare to fail to graduate school in South Korea. Therefore, we use the actual income amount in the KLIPS when analyzing only employed women. However, when analyzing a more comprehensive sample including nonworking women, regression may result in bias if the income of unemployed women is set to zero or excluded. As a result, the study classifies women into six groups: noneconomic activity, unemployment,Footnote ⁷ and the first through fourth income quartiles,Footnote ⁸ and uses categorical dummies in the regression. We separate groups who are unintentionally unemployed and without economic activities because the backgrounds of both groups of women highly differ in terms of non-participation in the labor market. The term “unemployed” pertains to a person who fails to find a job despite the desire to work. By contrast, “non-economic activity” denotes a person who voluntarily does not work for lack of desire. Therefore, the study expects that the two groups will behave differently such that unemployed women are reluctant to give birth during the search for a job, whereas economically non-active women are not.

3.3 Kaplan–Meier failure function

Prior to the regression analysis, we draw the hazard functions by women's employment status by using Kaplan–Meier's nonparametric estimation (Figure 2). According to the results, women's employment is better for childbirth than involuntary unemployment.Footnote ⁹ In the second panel of Figure 2, which divides income levels into quartiles, the lowest risk of giving birth is in the first income quartile, the risk increases in the third and second quartiles, and the highest probability of giving birth is in the highest (fourth) quartile. These findings contrast with the traditional hypothesis that employment or higher incomes have a negative impact on childbirth in South Korea.

Figure 2.

Kaplan–Meier's nonparametric estimation.