Current Issue

Previous studies to identify factors affecting marriage include both unmarried and married people, but those looking into childbirth mostly considered married people as the subjects. If the targets are limited to married people, however, the sample selection bias is likely to increase as the sample is not randomly drawn but people whose marital status is single (‘not married’) are arbitrarily excluded. With this in mind, this study used the Heckman sample selection model to compensate for the sample selection bias.

The Heckman sample selection model assumes that the factors affecting marriage and the number of children (childbirth) are different, and hence the factors determining marital status and the number of children born can be analyzed in two stages. The selection equation (Eq. (1)) that estimates the variable (z) for marital status in Stage 1 and the regression equation (Eq. (2)) that estimates the variable (y) for the number of children in Stage 2, both of which were used by this study, can be expressed as follows (Park and Cho, 2016):

(1)

(2)

where w and x are independent variables, and β and γ are parameters to be estimated. μ and ε are the error terms, and the standard deviations of the error terms are 1 and σ, which are normally distributed (Min and Choi, 2021).

In Stage 1, the probit model is used to estimate the probability of sample selection, and in Stage 2, the OLS regression model (OLS) is applied to only the samples (z= 1) selected in Stage 1. The OLS is shown in Eq. (3) below (Greene, 2003; Lee and Lee, 2013):

(3)

In Equation (3), Φ(•) is the standard normal cumulative density function, and ϕ(•) is the standard normal probability density function. The inverse Mills ratio (IMR), the ratio of the two functions, is the probability that each observation will be momentarily excluded from the sample and can be expressed as λ(- w'γ). IMR is added as a new independent variable for the second stage analysis, and a significant value of the IMR indicates that the use of the Heckman sample selection model is valid. Also, ρσ(β_λ) serves as the regression coefficient and σ is the standard error of the regression model residuals, and ρ(rho) is the correlation between the residuals in the first and second analytical stages, with a value between -1 and 1. If the correlation coefficient of the residuals is high, the estimate of the OLS without IMR includes sample selection bias (Cho, 2017; Lim et al., 2018).

The empirical analysis utilized the Seoul Survey (2022) conducted by the Seoul Government. The Seoul Survey aims to analyze changes in Seoul, evaluate policy performance, and monitor changes in citizens’ quality of life, attitudes, and values in order to formulate relevant municipal projects. It is the most suitable data for the purpose of this study because it contains information on the marital status and children of households living in Seoul, as well as household head characteristics, economic characteristics, residential characteristics, and the residents’ perceptions of socio-economic risks living in the city and the possibility of upward mobility for themselves and their children.

The age range for the analysis is 20-44 years old, which is the range utilized in previous studies (Statistics Korea, 2013; Lim et al., 2018), where marriage, childbirth, and child-rearing are major events in a family’s life cycle. In the empirical analysis, we included 5,574 households out of a total of 20,000 households whose head of household age met the analytical criteria, excluding non-response and missing data. Of these, 1,712 are single and 3,862 are married households.

For the dependent variable using the Heckman sample selection model, we used ‘marital status’ for the first stage and ‘number of children’ for the second stage analysis (See Table 1). In Stage 1, we categorized all households into single (0) and married (1), with 69.3% of households being married. In Stage 2, the number of children in married households was set as a dependent variable, with a minimum of 0 and a maximum of 3, and a mean of 0.3.

The independent variables were divided into socio-economic characteristics, residential characteristics, perceived risk of socio-economic situation, and upward mobility likelihood of family members (which are believed to affect marriage and number of children based on previous studies), and the independent variables for the first and second stage analyses were the same. First, the age and education of the household head, which are socioeconomic characteristics, were found to be 36.7 years old on average and university graduate or lower, respectively. Household income, which represents the economic power of the household, was 4 to 6 million won or lower on average, and 55.2% of the households had household debt.

Next, residential characteristics included housing type, occupancy type, and residential areas. The housing type was divided into apartment, detached house, and multi-family house, and the occupancy type was divided into home owners, jeonse (lump-sum housing lease), and rent, with apartment and rent serving as the reference variables. As for the housing types, apartment (41.3%) was the highest, followed by multi-family house (36.1%), and detached house (22.6%). In terms of occupancy type, jeonse was the highest with 56.4%, followed by rent (22.6%) and home owners (21.0%), indicating a higher proportion of renters (79.0%) than homeowners. Residential areas were divided into five living areas in Seoul, with the central area as the reference variable. Among the residential areas, the largest number of households lived in the northeast (35.0%) and the smallest in the central area (8.0%).

Risks that occur in cities can affect lives, and socioeconomic factors are a major variable in marriage and childbearing (Chun, 2013; Kim and Jeon, 2020). In this regard, the risk perception of the socioeconomic situation in the city considered in the study was assessed in terms of unemployment, economic crises (e.g. financial crises), and social conflicts (e.g. wealth gap, inequality, and generational conflicts); all three categories were perceived as slightly serious, with a score of 3.8 for unemployment and economic crises and 3.7 for social conflicts.

Finally, we investigated the perceived probability of family members’ upward mobility; as a key variable of this study, it means the perceived likelihood of socioeconomic status changes of the subject (parents) and his/her children. We used two questions for the subject and his/her children, respectively, by asking: “How likely or unlikely do you think it is that your (or your child’s) socioeconomic status will become higher in our society?”. We first collapsed the five-point scale responses to a three-point scale (low, moderate, high), and combined the scores to the responses for the likelihood of upward mobility for themselves and their children. With this method, we could draw a total of nine categories, and ‘Low’ for oneself and ‘Low’ for the children (OL-CL) was respectively used as the reference variable. It should be noted that marriage and childbirth are past events, while the likelihood of upward mobility is a current perception, and thus there may be a time difference in perception. Existing studies, however, have shown that subjective perceptions do not change significantly over time (Walsh et al., 1996; Na and Cha, 2010), and given that it is not uncommon for families to make a decision for marriage and childbirth based on the possibility of upward mobility, this study assumed that people had the same perceptions toward marriage and childbearing decisions in the past as they do today. Regarding the perception of upward mobility, the highest percentage of responses were ‘High-Oneself’ and ‘High-Children’ (OH-CH, 26.7%), followed by ‘Moderate-Oneself’ and ‘Moderate-Children’ (OM-CM, 23.8%), and ‘Low-Oneself’ and ‘Low-Children’ (OL-CL, 18.0%), indicating that the subjects generally perceive their own and their children’s upward mobility similarly. The two categories with the lowest percentages were found to be ‘High-Oneself’ and ‘Low-Children’ (OH-CL, 1.5%) and ‘Low-Oneself’ and ‘High-Children’ (OL-CH, 3.1%).

The analysis showed that the IMR(λ) value is statistically significant, indicating that sample selection bias occurs within the analysis with the OLS (See Table 2). Therefore, using the Heckman sample selection model was found to be valid.

In the first stage of the analysis designed to identify the factors that influence the perceived upward mobility of family members to marriage, it was found that all socioeconomic characteristics have a significant impact on the likelihood of getting married. Age is positively related to the likelihood of marriage, but age squared is negatively related to the likelihood of marriage, indicating that the likelihood of marriage increases up to a certain level and then decreases with age. This may be attributable to the fact that one’s attitude toward marriage changes or people tend to give up on getting married as they pass the ‘marriageable age’ with time (Lee, 2019). Higher education level had a negative impact on marriage, as a long period of education delays marriage and reduces its probability, as explained in previous studies (Kim and Sun, 2011; Do and Choi, 2018). Higher household income, which is a proxy for economic stability, increased the probability of marriage, consistent with previous studies (Lee and Choi, 2012; Lim et al., 2018). Household debt was found to have a positive effect on marriage; this cannot be interpreted to mean that household debt increases the likelihood of marriage, but rather that marriage creates situations that require married households to borrow money, for wedding preparations and getting a home for the family.

Regarding housing characteristics, living in a detached or multi-family houses rather than an apartment, and renting rather than owning or jeonse, were found to be negatively related to marriage. This is because when it comes to marriage, apartments are what people desire most because of the convenient living environment, and owning a home or Jeonse are preferred to monthly renting because of relatively higher residential stability. Also, people tend to delay marriage to secure their preferred living space (Kang and Ma, 2017; Byun et al., 2018), and thus the likelihood of being unmarried increases when living in places other than apartments or by monthly renting. These findings confirmed that convenient living conditions and residential stability are important factors in the decision about marriage (Lim et al., 2018; Lee and Seo, 2019).

The probability of marriage decreased when living in the central area compared to the northwest or southwest areas. High housing prices and the lack of a preferred housing environment for a family have a negative impact on marriage (Do and Choi, 2018; Lim et al., 2018). In this regard, the central area had relatively higher housing prices (both for purchasing and jeonse)³⁾ than the northwest and southwest areas, while having relatively lower ratio of apartments (preferred housing type for married couples) and after-school care centers and care facilities (factors considered when couples decide to have a baby after getting married)⁴⁾. These factors may explain the lower probability of marriage among residents of the central area.

Next, in terms of risk perception of socioeconomic factors, only social conflict was found to have a significant impact; the more severe the perceived problems such as wealth gap, inequality, or generational conflict are, the higher the likelihood of being unmarried is. Wealth gap or inequality that exists in society makes it difficult for individuals to take the general life course that starts from meeting a loved one to getting married and having a child only on the basis of one’s individual achievement and creates disparities in life course fulfillment by class (Oh, 2020). These aspects seemed to have influenced the above findings for perceived risks of socioeconomic conditions. In addition, generational conflict stemming from different perceptions of marriage, housework, and childbirth between generations can lead to family conflict, and this can be a negative factor for marriage. From this, it can be inferred that risks existing in cities such as social conflict are relevant to individuals’ lives, especially marriage, and hence mitigating social conflict can play a role in increasing the likelihood of marriage.

Finally, regarding the likelihood of upward mobility of family members, the key characteristic dealt with in this study, we found that ‘Low-Oneself’ and ‘Moderate-Children’ (OL-CM), ‘Moderate-Oneself’ and ‘Low- Children’ (OM-CL), and ‘Moderate-Oneself’ and ‘High-Children’ (OM-CH) categories had a significant impact. Compared to the reference variable ‘Low-Oneself’ and ‘Low-Children’ (OL-CL), the probability of marriage decreases for ‘Moderate-Oneself’ and ‘Low-Children’ (OM-CL) category. This suggests that the less optimistic people are about their children’s future than their own, the more negative they are about marriage, and this had a greater impact than perceiving the entire family members (including oneself and their children) as less likely to move up the social ladder. Conversely, the more they expected their children to be more likely to move upwardly the social class than themselves (OL-CM and OM-CH), the more likely they were to decide to marry, even if they did not perceive their and their children’s probability to move upward as high. These findings suggest that in order to mitigate non-marriage and late marriages that eventually lead to not having children or delayed childbearing, and thus a serious low fertility rate, it is necessary to provide people with an optimistic outlook that their offspring will have a higher socioeconomic status than the parents themselves.

With the results from the first stage analysis examining the impact of family members’ upward mobility on marriage, in the second stage analysis we also examined the effect of family members’ upward mobility on the number of children born. We found that age and household debt were significant socioeconomic characteristics, with age being associated with an increase in the number of children born up to a certain age, similar to the biological fertility range, and then decreasing thereafter (Song, 2014). Household debt was found to be positively related to having more children. It appears that a married couple, in general, needs to borrow money to finance the increased expenses for childcare and housing after delivery (e.g. lump-sum renting (jeonse) or home owning, improving the living environment, etc.) (Kim, H.S., 2017).

In terms of housing and occupancy types included in the housing characteristics, apartments with relatively better education and living conditions, and jeonse or owning a home with more stable housing environment than renting, were positively associated with fertility and multi-child birth, in line with previous studies (Kim, K.A., 2017; Lee and Noh, 2017; Lee and Seo, 2021). This suggests that housing stability and living conditions are important factors for childbearing as well, just as they are for marriage. In terms of residential areas, the number of children born was found to be higher in the southeast and northwest regions⁵⁾ than in the central area. Given that decent childcare infrastructure can effectively increase regional fertility rate (Park and Jang, 2022), the likelihood of having a child is higher in the southeast and northwest regions equipped with childcare infrastructure such as regional offices of community child centers, after-school care classes, and Together Care Centers (government-run after-school care facilities), etc. than in the central region. From these results, this study concludes that expanding local after-school care and childcare infrastructure is important to improve fertility rates.

Interestingly, despite that the subjects perceived living in a city was risky, with the scores of unemployment, economic crisis, and social conflict being 3.6 or higher on average, none of the variables had a significant effect on fertility. This suggests that the decision to give birth may be largely determined by personal circumstances rather than macro socio-economic situations.

Finally, the likelihood of upward mobility of family members was a major driver in increasing the number of children born in the ‘Moderate-Oneself’ and ‘High-Children’ (OM-CH) and ‘High-Oneself’ and ‘High-Children’ (OH-CH) categories. In other words, respondents were more likely to have multiple children when they perceived they are likely to climb up the social ladder and their children are highly likely to be upwardly mobile. Regarding marriage, the respondents answered that one can marry if his/her children’s likelihood of upward mobility is high (even if their own probability of upward mobility is low), as examined in the Stage 1 analysis, but childbearing was found to be more likely to occur when one’s upward mobility is at least moderate and one’s children’s upward mobility is high. This suggests that people need a much higher probability of family upward mobility to decide to have a child than they do to decide to marry, and that the low fertility problem can be resolved by creating an environment where people can expect their children to have a high probability of upward mobility.