South Korea and other developed regions in East Asia have become forerunners of prolonged lowest-low fertility. South Korea's total fertility rate has been below 1.3 for two decades, the longest duration among OECD countries. Using vital statistics and census data, I study recent trends in the country's cohort fertility covering women born before the 1960s to those born in the 1980s. Analyzing outcomes at both the intensive margin of fertility (i.e., timing and number of children) and the extensive margin of family formation (i.e., marriage and childlessness), I document three novel patterns. First, the driver of low fertility has evolved across birth cohorts, from married women having later and fewer childbirths, to fewer women getting married, and finally to fewer women having children even if married. Second, a decomposition analysis of marriage and fertility changes indicates that the marriage and fertility decline was driven by changes within educational groups rather than by changes in women's educational composition. Third, the relationship between women's educational attainment and marriage or fertility was negative for the 1960s cohort, but an inverted U-shaped education gradient emerged beginning with the 1970s cohort.
In the 1990s, the total fertility rate (TFR) fell below 1.3 children per woman in parts of Europe and East Asia (see online appendix Figure A1, panel a).1 The phenomenon alarmed policymakers because prolonged periods of “lowest-low fertility” (coined by Kohler et al. 2002) would result in rapid population decline. Although researchers agreed that the emergence of very low TFR was due to fertility postponement, they had different views about how long lowest-low fertility would last. Some studies predicted the phenomenon to be transient, given that delays in childbearing eventually cease (Bongaarts 2001, 2002; Sobotka 2004). Others provided a cultural explanation for why it could continue (Feyrer et al. 2008; McDonald 2000). The turnaround in TFR in many low-fertility countries in the 2000s, however, implied that the decline in period fertility measures was mostly due to women delaying childbirth rather than having fewer children. The era of lowest-low fertility seemed to come to an end, at least in Europe (Goldstein et al. 2009).
By contrast, lowest-low fertility in South Korea (hereafter, Korea) has been ongoing since 2001. In fact, the TFR has fallen even more in recent years and hit a world record low of 0.84 in 2020. Other developed regions in East Asia (e.g., Singapore, Taiwan, and Hong Kong) are also experiencing prolonged periods of very low fertility (see Figure A1, panel b) despite years of pronatalist policies (Frejka et al. 2010; Jones 2019).2 East Asia reopens the debate on the prospect of lowest-low fertility because it can no longer be explained by tempo effects and shows no signs of reversal. Korea may become the first OECD country to resemble the “low-fertility trap” (Lutz et al. 2006), in which self-reinforcing mechanisms result in a continued fertility decline.
In this article, I describe recent trends in Korea's cohort fertility to elucidate the development of prolonged lowest-low fertility. To investigate whether the main driver of low fertility has evolved over time, I study changes at the intensive margin of fertility (i.e., timing and number of children) and the extensive margin of family formation (i.e., marriage and childlessness). I also explore the role of women's educational attainment in explaining the differences in these outcomes across and within cohorts to provide insight into potential mechanisms. I use the latest available data from vital statistics (1981–2021) and the census (1966–2020), covering women born before the 1960s to those born in the 1980s.
By using a cohort approach, the study has several advantages over prior research on low fertility focusing on period measures. Most notably, period measures, such as the TFR, can fluctuate from transitory changes in fertility timing even when the total number of children does not change (i.e., tempo effect). Such distortions can be significant when women's childbearing age shifts across cohorts, as in Korea (Bongaarts and Feeney 1998; Kye 2012). Using data on women aged 40 and older, I calculate each cohort's number of children and the share of women never married or childless rather than adjusting period measures to account for tempo effects.3
Even without significant tempo effects, cohort outcomes also have the advantage of being straightforward to interpret. Because the TFR is synthetic, it does not correspond to the fertility behavior of an actual group of women. A low TFR can be derived from a higher order birth decrease among older women or a first-childbirth decrease among younger women. By examining each cohort's outcomes, I provide useful evidence for policy design and enable easier comparisons across countries.
Moreover, cohort analysis can deepen our understanding of fertility changes by situating them in the context of each generation's socioeconomic experiences. As Goldin (2004) demonstrated, the set of constraints and opportunities each cohort experiences can explain their behaviors. Such an attempt would be particularly helpful for examining Korea, where rapid economic growth led to vastly different experiences across generations.
I document three key patterns. First, the driver of low fertility has shifted across cohorts, from married women having later and fewer childbirths, to fewer women ever marrying, and finally to fewer women having children even if married. That is, significant changes have occurred not only at the intensive margin of fertility but increasingly at the extensive margin of family formation, with a clear decoupling of marriage and childbearing beginning with the late-1970s cohort.
Second, a decomposition analysis of marriage and fertility changes by women's educational level indicates that the marriage and fertility decline was driven more by changes within educational groups than by changes in educational composition. Although women's educational attainment increased substantially over time, the shift in educational composition explains less than 8% of the decline in the average number of children between the 1956–1960 and 1976–1980 cohorts.
Third, the relationship between women's educational attainment and marriage or fertility has changed from negative to an inverted U-shape. For the 1960s cohort, better educated women were less likely to marry or have children. For the early-1970s cohort, however, the education gradient is slightly nonlinear, with marriage and fertility rates of women without a high school degree falling below those of women who are high school graduates. The pattern is more evident for the late-1970s cohort: women with the highest (graduate school degree) and lowest (less than a high school degree) educational levels have lower marriage and fertility rates than women who are high school or college graduates.
Korea was one of the world's fastest-growing economies in the late twentieth century, transforming from a developing country to a developed one within a few generations. In this section, I provide a brief overview of changes in Korea's socioeconomic and cultural factors relevant to the study period.
Education and Markets
Rapid economic growth in Korea was accompanied by a remarkable expansion in higher education. Only 13% of women born in 1960 but 51% of women born in 1985 are four-year college graduates. The expansion occurred for both men and women, erasing the gender gap in tertiary education. In fact, Korea has the highest proportion of 25–34-year-olds with tertiary education among all OECD countries (OECD 2021a).
As more women became highly educated, more entered the workforce and with higher earnings potential. For the birth cohorts of 1960 versus 1985 at ages 25–34, the female labor force participation rate increased from 44% to 68%, and the share of women with managerial or professional occupations more than tripled (increasing from 8% to 28%).4 The gender wage gap, defined using median earnings among full-time employees, decreased from 52.8% in 1985 to 32.5% in 2019 (OECD 2021a).
Around the time when the early-1970s cohort entered the labor market, however, Korea experienced its worst recession since the Korean War: the 1997 Asian financial crisis. Many firms filed for bankruptcy and laid off workers, and the crisis had lasting impacts on Korea's labor market structure. Job precariousness increased, with the number of temporary workers increasing from 16.6% of total wage and salary employment in 2001 to 28.3% in 2021, the second-highest rate among all OECD countries (Grubb et al. 2007; OECD 2022). The youth unemployment rate also increased, from 5.5% in 1990 to 9% in 2020.
The economic slowdown and greater labor market uncertainty intensified competition for elite university admission and stable jobs, increasing the educational cost of raising children (Anderson and Kohler 2013; Kim et al. 2021). For instance, 75% of students in elementary, middle, and high school received private out-of-school education (e.g., cram school, private tutoring) in 2021, and the average monthly expenditure per student increased from $230 in 2007 to $292 in 2021 (Ministry of Education 2010, 2022).5 Among married households with the oldest child aged 6–17 in 2019, the share of household expenditure on education was 14.1% in Korea compared with 3.3% in the United States (U.S. Bureau of Labor Statistics 2013–2020).6 The proportion of spending on tertiary education funded by households is also one of the highest among OECD countries, at 41.4% in 2018; this rate is similar to that in the United States (44.6%; OECD 2020b).
Moreover, housing prices increased substantially. During the past decade, the housing price-to-income ratio rose from 3.9 to 5.1 times in cities and from 8.6 to 12.9 times in the capital city, Seoul (Korea Real Estate Board 2022).7 The market is also rigid because the common type of lease in Korea is not a monthly rent but rather a large lump-sum deposit (jeonse), typically more than half the market value.8
In sum, the opportunity and actual costs of children have increased. On the one hand, the rise in women's earnings potential has made time away from work for childcare more costly. On the other hand, with lower prospects for lifelong jobs and higher costs of raising children and securing a home, the younger generation does not believe that they can afford to start a family (Lim 2021). A negative income effect dampens the demand for children (Becker 1991).
In the 25 years since the adoption of the National Family Planning Program in 1961, Korea's TFR fell from 6 to 2. As shown in Figure A2 (panel a), higher order births fell dramatically, and two-child families became the norm beginning with the 1950s cohort.9 With Korea reaching subreplacement fertility, policy emphasis in 1996 shifted from population control to population quality and welfare.
New policies addressing low fertility and population aging were enacted in the early 2000s. The First Basic Plan on Low Fertility and Aged Society was initiated in 2006, with follow-up plans in 2010 and 2016. The government spent more than $76 billion on this program, which includes the extension of paid parental leave, expansion of early childhood education and care, and introduction of universal child allowance for young children (Lee 2018). As a result, public spending on families has grown more than tenfold since the early 2000s, and early childhood education and care service enrollment rates for children ages 0–2 and 3–5 have exceeded the OECD average (OECD 2019).
From an international standpoint, however, Korea's public family spending still remains low relative to that in other developed countries. In 2017, this spending was 1.1% of GDP compared with the OECD average of 2.1% (OECD 2019). Short primary school hours relative to parents' long working hours result in gaps in childcare. Further, Korea's culture limits policy implementation, as described in the next subsection.
Work Culture and Gender Norms
As a legacy of the labor-intensive high-growth era, Korea has an intense work culture. Korea has the fourth-longest working hours among OECD countries. Overtime work is frequent, and employers highly value workers' loyalty and commitment (Brinton and Oh 2019; OECD 2021b). Hence, workers find it difficult to use policies such as parental leave without concerns about discrimination in the workplace (Moon and Shin 2018).
Meanwhile, as a legacy of Confucianism, Korea has patriarchal, traditional gender norms: men are assumed to be the main breadwinner, and women are assumed to be the main caretaker. According to the 2017–2018 World Value Survey, more than half of Koreans agree with the statements “Men should have more right to a job than women” and “[A] pre-school child suffers with working mother” (53% and 65%, respectively). By contrast, in the United States, only 5% and 21%, respectively, agree with these statements. Time-use data also depict traditional gender roles (Hwang et al. 2019). Among dual-earner couples with a young child, mothers spend an average of 3.6 hours more per day on housework and childcare than fathers in Korea (Choi 2018) compared with 1.2 hours more in the United States (U.S. Bureau of Labor Statistics 2020).10
These cultural features and the previously noted economic and policy factors can make balancing work and family very difficult in Korea. Many women work to make a living or to pursue a career. However, the workplace is not family-friendly, and the household is not work-friendly for mothers.11 Public support for families has expanded, but it remains insufficient to address these dual responsibilities of work and childcare. In Korea's 2020 Local Area Labour Force Survey, 43% of married women who quit work reported childcare as the reason. The female labor force participation rate continues to dip around women's childbearing age (Figure A3).
Data and Methods
I use administrative records and nationally representative survey data to comprehensively describe cohort fertility trends in Korea. First, I use 1981–2020 birth and marriage registers from the Korean Vital Statistics. Birth records include information on parents' age, marital status, and mother's total number of children for all births registered in Korea. Marriage records include information on the husband's and wife's age and marriage type (e.g., first marriage).
Because vital statistics, by construction, include only those women who married or gave birth, I combine these data with population data to calculate marriage and fertility rates. To obtain the female population by age and birth year, I use data from population projections (Projected Population) and population statistics based on resident registration (Resident Population). Both are official statistics published by Statistics Korea, but they differ in years and definition of population. Projected Population is available from 1960 to 2067, with population estimates from 1960 to 2017 as of July 1 each year. The population here includes all individuals residing in Korea at the time, regardless of nationality. Resident Population, on the other hand, reports the number of residents as of July 1 in each year excluding foreigners; these data are available for 1993–2020.12 However, the share of foreign residents in Korea is minimal, at only 3.3% in 2020 (Statistics Korea 2021).
These merged data can be used to calculate the age-specific fertility rate, the cumulative cohort fertility rate, and the share of women never married or childless for each birth cohort. The age-specific fertility rate (ASFR) and the cumulative cohort fertility (CCF) for five-year cohort group c are defined as follows:
where b is the first birth year of cohort group c, birthx,a is the number of childbirths to women born in year x at age a, and popx,a is the total number of women born in year x at age a. ASFRc,a is the number of births to women of cohort group c at each age a per 1,000 women in that cohort group and age. CCFc,a is the average number of children born to a woman belonging to cohort group c, from age 15 to age a. Unlike the TFR, which involves underlying assumptions about tempo, ASFR and CCF are actual numbers of births to women in each cohort.
The percentage of women in birth cohort c who were never married by age 40 is calculated as follows:
where marriedc,a is the total number of women in cohort c with a first marriage between ages 15 and 40 in the marriage register, and popc,40 is the total number of women in cohort c at age 40 in the population data. Because vital statistics are available beginning in 1981, I begin the analysis of these data with the 1966 birth cohort so that I can count all women who married starting at age 15.13 The percentage childless is calculated analogously using birth registers and population data.
I use age 40 as the threshold because it is old enough to define women's first marriage and childbirth while young enough to obtain data for recent cohorts. Reproductive life can extend to age 50 for some women. However, extensive margin measures, such as whether women are never married or childless, are determined by age 40 for the vast majority of women. Even in the most recent vital statistics (collected in 2020), for instance, 97% of first marriages and 98% of first childbirths occurred to women aged 40 or younger. Outcomes at the intensive margin of fertility, such as the total number of children, would also be similar to complete fertility, given that women face increasing difficulty becoming pregnant as they approach age 40.
I also use the Korean Population and Housing Census, which Statistics Korea conducts every five years. I use all available years of the 1966–2020 1% sample microdata. The data contain information on individual-level characteristics, such as age, sex, educational attainment, marital status, and the number of children.
The census has two advantages over vital statistics. First, the census covers all women regardless of marital status or fertility. I can thus use census data to calculate, for each cohort, the average number of children born and the share never married or childless at age 40 without concerns about data discrepancies.14 Second, using information on demographic characteristics allows me to study marriage and fertility outcomes by women's educational level. A caveat of the census is that data on fertility are collected only for ever-married women. This limitation is not much of a concern in Korea, however, because out-of-wedlock childbirths are very rare (less than 2% of all births). As in prior studies using the census, I assume that never-married women are childless (Choe and Retherford 2009; Yoo 2014).
Cohort Trends in Marriage and Fertility
Displaying changes at the intensive margin, Figure 1 plots the ASFR (panel a) and the CCF (panel b) for each cohort group, from 1961–1965 to 1981–1985, from vital statistics. Significant differences are evident across cohorts in childbirth timing (tempo effect) and the number of children born per woman (quantum effect). Panel a shows that the ASFR was highest among women aged 26 in the 1961–1965 cohort but among women aged 31 in the 1981–1985 cohort. Panel b shows that the average number of children per woman declined from 2 to 1.2 across cohorts. Although the youngest cohort's fertility can be observed only until age 39, the flattening CCF curve suggests that women's fertility may be nearly complete.
The trend of women having fewer children and at later ages has not been gradual across cohorts. For the 1961–1965 and 1966–1970 birth cohorts, the shape of the ASFR distribution changed slightly as fewer women had children in their early 20s, but the average number of children per woman did not change cumulatively. For the 1970s birth cohorts, however, the entire distribution shifted right as the ASFR dropped substantially among women in their 20s but rose among women in their mid-30s. The average number of children per woman also dropped without recuperation at older ages, to 1.7 in the 1971–1975 cohort and then to 1.5 in the 1976–1980 cohort. That is, only a slight tempo effect is evident among women born in the 1960s, whereas significant tempo and quantum effects kick in for women born in the 1970s. In the most recent cohort, 1981–1985, additional tempo effects seem limited, but cumulative fertility fell even further to 1.2 children. A decomposition of parity progression ratios indicates that the decrease in cohort fertility since the 1960s cohort is mostly due to the decline in the transition to first and second births rather than to changes in higher order births (Figure A2, panel b).
Next, I examine changes at the extensive margin of family formation: whether a woman was ever married and whether she had any children. Figure 2 plots the percentage of women never married or childless at age 40, assessed using both data sources. The percentage never married is close to zero, and the percentage childless is lower than 4% among women born in the 1930s, 1940s, and 1950s, confirming near-universal marriage and childbirth up through the 1950s cohort. A slight departure from this pattern emerges beginning with the mid-1960s cohort: among women born in 1970, 7% were never-married, and 10% were childless at age 40 in the census (6% and 9%, respectively, in the vital statistics data). Thereafter, both never-married and childless rates increased, the latter at a steeper pace. By the 1980 birth cohort, nearly 18% of women were never married, and 26% of women were childless at age 40 in the census (14% and 22%, respectively, in the vital statistics data). Both data sources reveal a higher rate of childlessness than in the United States, where 15% of women are childless at ages 40–44 (Current Population Survey 2018).
In addition to the surge in the share of women never married or childless, the widening gap between the two is noteworthy. As in other East Asian countries, marriage was considered a precursor to childbirth in Korea. Figure 2 indicates that this is no longer the case, not because more women have children out-of-wedlock but because more women do not have children even after marriage. The share of married women who are childless remained at approximately 4% until the 1970 cohort but more than doubled, exceeding 9%, for the 1980 cohort.
In sum, the decades-long low fertility in Korea is characterized by changes at both the intensive and extensive margins of fertility. Women born in the 1960s delayed childbearing to their late 20s, but most married and eventually had an average of two children. Childbearing for women in the early-1970s birth cohort is characterized as “later and fewer”; recuperation at older ages was limited, and cohort fertility fell to 1.7 children. Never-married and childless rates also began to rise, departing from the once near-universal pattern. Beginning with the late-1970s birth cohort, women further delayed childbirth to their 30s, cohort fertility fell below 1.5 children per woman, and major changes emerged at the extensive margin as the link between marriage and childbearing severed and increasingly more women (even married women) remained childless.
Differential Trends by Women's Education
Educational attainment is a proxy for one's potential income in the labor market and thus also captures the opportunity cost of time away from work for child-rearing. To examine the relationship between women's education and fertility, I divide women into four educational groups: less than high school graduate, high school graduate, four-year college graduate, and graduate school degree. I use 2000 census data because it is the earliest survey year that distinguishes between “some college” or a junior college degree from a four-year college degree. I cannot use vital statistics for the analysis because the population data (the denominator in Eqs. (1)–(3)) are not available by educational group. To plot yearly outcomes, I therefore present observations at age 40 (the survey year) and at ages 41–44 (in between survey years) from the census. Doing so allows me to compare birth cohorts from 1956–1960 to 1976–1980. Across these cohorts, the share of women who did not complete high school fell from 39% to just 2%, whereas the share with a college degree increased from 8% to 34% (Figure A4).
Considering the remarkable rise in women's education, I first investigate how much of the change in family outcomes is attributable to the change in women's educational composition across cohorts. Specifically, I calculate the average number of children, the percentage never married, and the percentage childless among women aged 40–44 in each educational group and birth cohort cell. Following Das Gupta (1993), I then decompose the total change in outcome between two cohorts into changes within an educational group (the rate effect) versus changes in educational composition (the compositional effect) by using the average over the two populations as the standard.
Table 1 reports the result from the decomposition analysis between the 1956–1960 and 1976–1980 cohorts.15 Panel A presents the rate effect and composition effect of the change in the average number of children per woman. The total change between these cohorts is 0.5 fewer children per woman, 92% of which is due to the rate effect and only 8% of which is due to the compositional effect.16 That is, almost all the decline in the average number of children across cohorts is due to the trend within each educational group and not because more women obtained a college degree over time. Similar results are shown in panels B and C, which decompose the change in the share of women never married and childless, respectively. Of the total increase in the share of single women, 97% is due to the rate effect. For childlessness, the rate effect exceeds 100%, meaning that childless rates would have been even higher if not for the change in educational composition.17 The large rate effect in all three outcomes confirms a strong cohort trend of declining marriage and fertility regardless of women's educational level, consistent with prior studies (e.g., Choe and Retherford 2009; Jones and Gubhaju 2009; Shin 2019).18
Next, I study the education gradient in these outcomes within each cohort. Figure 3 plots the average number of children at ages 40–44 by education and birth year among all women (panel a) and ever-married women (panel b). The average number of births per woman declined significantly at all levels of education, consistent with the large rate effect found earlier. The cohort trend is similar in panels a and b, although the average number of children is higher when I restrict the sample to ever-married women.
Within each cohort, I find a negative relationship between women's education and the average number of children throughout the late-1950s and 1960s cohorts. On average, better educated women had fewer children. Beginning with the early-1970s cohort, however, fertility is slightly lower among those who did not complete high school (1.63) than among high school graduates (1.69). The difference between the two educational groups widens among the late-1970s cohort, such that the education gradient becomes a more inverted U-shape. The average number of children is highest among high school graduates (1.54), followed by college graduates (1.41) and then by women in the lowest (1.27) and highest (1.2) educational groups. Again, because I use observations at ages 40–44, differential timing due to schooling does not affect the results.
Figures 4 and 5 plot outcomes at the extensive margin, the share of women who remain single and childless by educational level, respectively. Figure 4 shows an upward shift in nonmarriage rates across cohorts regardless of women's educational level. The education gradient in marriage is negative until the early-1970s cohort, for which I begin to observe a switch at the lower end of the education spectrum as never-married rates jump among less than high school graduates. For most women, however, higher educational attainment remains associated with lower marriage probability: even in the 1976–1980 cohort, the never-married rate is 13% among high school graduates but 15% among college graduates and 22% among women with a graduate school degree. Such a relationship between women's education and marriage is in sharp contrast to the positive education gradient among men in Korea (Figure A5).19
Figure 5 presents the percentage childless at ages 40–44 by women's educational level. Panel a shows that the increase in childlessness was not large until a jump of nearly 10 percentage points at all educational levels between the two consecutive cohorts of 1966–1970 and 1971–1975. At the same time, the education gradient in childlessness clearly switched from negative to U-shaped. Beginning with the early-1970s cohort, and more evidently in the late-1970s cohort, women in the lowest (less than high school graduate) and highest (graduate school degree) educational groups had higher childlessness rates (30% and 29%, respectively) than women with intermediate levels of education (20% among high school graduates and 22% among college graduates).
In the sample restricted to ever-married women in panel b, the education gradient in childlessness changes from being nearly flat to U-shaped. Until the 1960s cohort, childless rates among ever-married women remained below 5% regardless of educational level because most women had at least one child after marriage. That is, an education gradient existed in marriage (Figure 4) but not in whether to have a child conditional on marriage. Beginning with the early-1970s cohort, however, childlessness rates were higher even among married women, particularly owing to the jump in childlessness among those with the lowest and highest levels of education.
In sum, there are two main findings from the education analysis. First, cohort trends in declining marriage and fertility appear across all educational groups such that the over-time change in women's educational composition per se cannot explain most of the decline over the past few decades. Second, within each cohort, the relationship between women's educational attainment and family outcomes was negative throughout the 1950s and 1960s cohorts, but an inverted U-shaped relationship emerged in the 1970s cohort because marriage and fertility decreased further among women who did not graduate high school than among those who did.
This study analyzes recent cohort fertility trends in South Korea, a country experiencing the longest duration of lowest-low fertility among OECD countries. The study is the first to document trends at both the intensive margin of fertility (i.e., the timing and number of children) and the extensive margin of family formation (i.e., marriage and childlessness) across cohorts of women born in the 1960s to the 1980s. My analysis by women's educational level also allows me to explore potential mechanisms underlying Korea's low-fertility era.
Three phases seem to emerge from the study. First, women born in the 1960s delayed childbirth, but most married and had approximately two children by the end of their reproductive age. The education gradients in marriage and fertility were negative: women with higher educational attainment were less likely to be married and to have a child, and they had fewer children, on average. The early-1970s cohort marks the second, transitional phase. Childbirth was postponed further without recuperation such that cohort fertility fell to 1.7 children per woman. Never-married and childless rates began to rise. Education gradients became slightly nonlinear as marriage and fertility rates among those without a high school degree fell below those of high school graduates. In the last phase, fertility dropped further to 1.5 children for the late-1970s cohort and then again to 1.2 for the early-1980s cohort. The near equivalence of marriage and childbirth diverged as the share of childless women increased even among the married. An inverted U-shaped education gradient became clearer, with the lowest (less than high school graduate) and highest (graduate school degree) educational groups having lower marriage and fertility rates than intermediate (high school and college graduate) groups.
Although identifying causal effects is outside the scope of this study, potential channels for my findings are worthy of discussion. The first phase portrays an extension of the prior family planning era. Women born in the 1960s were in their 30s between the end of the National Family Planning Program and the beginning of pronatalist policies. Most women continued to get married and follow the two-child norm. Many of these women, even those who were college educated, did not work after marriage. In fact, the probability that married women worked in their 40s was negatively associated with education except for those with a graduate school degree (Figure A6, panel b).
Why I observe distinct patterns from the 1970s cohort is an interesting question for further research, but at least three important developments can be considered. First, more women probably needed to work to support themselves and their families because the 1970s cohort entered the labor market following the 1997 Asian financial crisis. The rapid rise in precarious employment and increasing child-rearing and housing costs would have made it difficult for the younger generation to start a family (Adserà 2004). Those with fewer economic resources would have been particularly affected, consistent with my finding of sharp drops in marriage and fertility among the least educated.20
At the same time, the number of women who wanted to work probably also increased as more women began to have “careers” rather than “jobs.” Although the female college enrollment rate has continuously risen, women's share in male-dominated college majors grew substantially from the 1970s cohort (Figure A7). Within a decade, the number of women who entered professional occupations, such as law and foreign service, doubled.21 Among married women aged 40–44, the share of dual earners surpassed 50%, and the relationship between education and the probability of working became positive (Figure A6). The disjuncture between women's earnings potential and their traditional role as mother and wife became salient (Feyrer et al. 2008; Hwang 2016). Some highly educated women delayed marriage or chose not to marry when marriage implies child-rearing and a significant career interruption, the so-called marriage package (Rindfuss et al. 2009).22 Consistent with my finding of increasing never-married rates, the early 1970s cohort is the one for which the term “gold miss” was coined and widely used in Korea to refer to unmarried women with high socioeconomic status.23 The inverted U-shaped education gradients in marriage and fertility may thus reflect these two separate mechanisms: women may be childless because of poverty or the high opportunity cost of parenting (Baudin et al. 2015).
Lastly, feedback effects between demographic trends and attitudes toward family life could have accelerated the fertility decline. Despite the increase in childlessness, the decoupling of marriage and childbirth became evident among women born in the late-1970s. Compared with women in the previous cohort who may have chosen not to marry because of the ensuing marriage package, the new trend indicates that marriage and fertility have become separate decisions. This separation may be due to an increased cost of living or a shift in values. As Lutz et al. (2006) noted, a self-reinforcing process toward lower birth rates can occur when one's personal ideal family size is influenced by one’s peers and experiences. With more men and women remaining childless, the younger generation is more likely to value their leisure (with or without a partner) than to consider having a child as an essential part of life (Yi 2006).
Some of these Korean trends, such as the increase in childlessness (Ahn and Sánchez-Marcos 2020; Hellstrand et al. 2021) and its U-shaped relationship with women's education (Baudin et al. 2015; Zeman et al. 2018), are similar to those of Western developed countries. However, my findings also indicate distinct features of Korea. First, cohort fertility, not just the TFR, is clearly lower in Korea. Cohort fertility is approximately 1.9 for the 1975 cohort in Nordic countries (Hellstrand et al. 2021) and seems to be leveling off at approximately 1.5–1.6 for the 1970s cohort in most Southern European, German-speaking, and Eastern European countries (Jalovaara et al. 2019; Zeman et al. 2018). In contrast, no signs of stabilizing cohort fertility are evident in Korea: cohort fertility dropped further, to 1.2, for the early-1980s cohort.
Second, the education gradients in marriage and fertility changed in Korea—not because of a recovery in fertility among the highest educational group, but because of a drop among the lowest educational group. The relationship has historically been negative for women, but the education gradient vanished in Nordic countries beginning with the early-1970s cohort (Hellstrand et al. 2021) and weakened in the United States, France, and Germany as fertility increased for highly educated women (Doepke et al. 2022). The emerging inverted U-shaped pattern in Korea may thus imply persisting incompatibility between career and family for highly educated women and increasing economic hardship for less educated women. Several factors—including insufficient public support for young families, inflexible work culture, traditional gender norms, and labor market duality (Doepke et al. 2022; Goldin 2021; Grubb et al. 2007)—may be responsible. The trend also suggests that Korean women without a high school degree have become as selective as those with a graduate school degree, given that their population share diminished over time.
Lastly, the decoupling of marriage and childbirth in Korea occurred because of childless marriages, not nonmartial childbearing. In the United States and Europe, this link became weaker through a change in the temporal ordering of marriage and pregnancy or because cohabitation overtook marriage as a form of partnership: 40% of births in the United States and above 50% of births in France and Sweden occur outside of marriage (OECD 2020a). In Korea, extramarital births remain rare, at approximately 2%. Although childbirth need not follow marriage, the temporal ordering remains intact, distinct from the second demographic transition (Lesthaeghe 2010).
More broadly, my results indicate that lowest-low fertility can have very different trajectories across regions and that a seemingly monotonic decline in the TFR can mask substantial heterogeneity by family outcome (e.g., the number of children, marriage, childlessness), cohort, and educational level. The finding that more women are remaining single and childless also implies that minor changes in childcare subsidies or parental leave will not have meaningful effects in societies with ongoing lowest-low fertility. Rather, policy emphasis should be on significantly reducing the high fixed cost—both actual and opportunity—of having a first child as the younger generation is increasingly giving up or opting out of “family.”
I thank Daeil Kim, Jinyoung Kim, Seok Ki Kim, Eunhye Kwak, Chulhee Lee, Jungmin Lee, Munseob Lee, Soyoung Lee, and participants at the 2021 Asian and Australasian Society of Labour Economics Conference, the 2022 Korea Empirical Applied Microeconomics Conference, SNU-UCSD joint webinar, SNU Institute for Research in Finance and Economics seminar, Bank of Korea Economic Research Institute forum, Korea Institute of Finance seminar, and the Korea Inequality Research Network seminar for helpful comments. Sujin Lee, Kang Suk Lee, and Yesol Lee provided excellent research assistance. Financial support from the Institute for Research in Finance and Economics of Seoul National University, the New Faculty Startup Fund at Seoul National University, and the Creative-Pioneering Researchers Program through Seoul National University are gratefully acknowledged.
Throughout the article, figures designated with an “A” appear in the online appendix.
Urban areas in China have also recently observed very low fertility rates (Zhao et al. 2017). Japan has low fertility rates, but its TFR has been above 1.3 except in 2003–2005 (Rindfuss et al. 2009).
Alternative suggested methodologies to address the limitation of period analysis include tempo- and parity-adjusted TFR (Bongaarts and Feeney 1998; Bongaarts and Sobotka 2012) and average cohort fertility (Schoen 2004). The Bongaarts–Feeney approach requires fairly strong assumptions about the nature of postponement: for instance, in a given period, all age groups postpone births by an identical amount (Goldstein et al. 2009).
Author’s calculations using data from the 1981–2020 Economically Active Population Survey.
I use constant 2020 KRW and then convert to USD.
Author’s calculations using Korea’s Consumer Expenditure Survey to match the definition of the U.S. Bureau of Labor Statistics. Education spending here includes private education and tuition.
The price-to-income ratio is the nominal house price index divided by household income.
In 2014, it was reported that the average cost of jeonse in Seoul equaled $300,000. In this system, the owner makes a profit from reinvesting the jeonse deposit, instead of receiving a monthly rent. The tenant is allowed to stay in the property “rent-free” and is not required to make monthly payments for the period of the lease, which is usually two years. Utilities and other costs are paid by the tenant. Further description is available at https://en.wikipedia.org/wiki/Jeonse.
Several studies have documented Korea’s earlier fertility transition toward the replacement rate (see, e.g., Choe and Park 2006; Choe and Retherford 2009; Frejka et al. 2010; Yoo 2014).
Young child is defined as a child under age 7 (before first grade) in Korea and under age 6 in the United States.
Similar problems have been documented in other developed regions in East Asia, such as Japan, Singapore, Taiwan, Hong Kong, and urban areas in China (Boling 2008; Jones 2019; Jones and Gubhaju 2009; Yingchun and Zhenzhen 2020).
Foreign-born residents who have changed their nationality to Korea are included in the Resident Population. Information on nationality is not available in the vital statistics for earlier years (before 1991 in marriage registers and 2008 in birth registers). Therefore, I cannot restrict the sample to Korean citizens.
Korea’s legal age of consent to marriage is 18. Women older than 16 may marry with their parents’ consent.
The merged data, for instance, can be biased because of an observation time lag (immigration or death before age 40) or a difference in population definition (foreigners) between the vital statistics and population data.
I find qualitatively similar results when I conduct the analysis for the 1956–1960 and 1971–1975 cohorts or the 1961–1965 and 1976–1980 cohorts.
Some compositional effects in Table 1 have a negative sign. The compositional effect of educational group e between two cohorts pre and post is compe = (we,post – we,pre) 0.5(ye,pre + ye,post), where w is the population share, and y is the outcome of interest. The contribution of the compositional effect to explaining the total change in y is compe /(ypost – ypre) 100 (%). Thus, if the population share of e changed in the opposite direction of y, the effect would be negative.
Childlessness increased substantially among those without a high school degree, but their population share diminished across cohorts (see Figure 5).
Even if the composition effect itself is small, indirect spillovers from educational expansion may have contributed to the cohort trend.
Using information on first marriages in the vital statistics and categorizing education into three groups (less than high school graduate, high school graduate, college graduate or more) across birth cohorts from the 1960s to the early 1980s, I find that most couples have equal educational attainment (68% to 77%), whereas hypergamy decreased (from 22% to 11%). See Park and Smits (2005) for trends in educational assortative mating in Korea. I do not show fertility results by men’s education because the census asks only ever-married women for their childbirth histories.
The incidence of temporary employment among wage and salary workers in Korea is more than twice as high for those without a high school degree (74%) as for college graduates (35%). Women are also overrepresented: 56% of temporary workers in 2021 were female (Ministry of Employment and Labor 2021).
From 2000 to 2009, the share of applicants passing the bar exam and the national foreign service exam who were female jumped from 19% to 36% and from 20% to 49%, respectively (Statistics Korea 2017). The average age of applicants passing the exams is 25–29, corresponding to the 1970s cohort.
Han et al. (2022) found that a gender-role ideology that incorporates both traditionalism and egalitarianism is associated with lower fertility intentions.
The same term does not exist for men because they show opposite marriage patterns; men with higher educational levels are more likely to marry (Figure A5).