1. Introduction
The literature has shown that girls in countries with strong son preferences, including China and India, are discriminated against, prenatally. For example, when a fetus is found to be a girl in India, mothers receive less prenatal care [Bharadwaj and Lakdawala (Reference Bharadwaj and Lakdawala2013)], and a female fetus is more likely to be aborted than a male fetus in both countries [Ebenstein (Reference Ebenstein2010), Chen et al. (Reference Chen, Li and Meng2013), González (Reference González2018), Bhalotra et al. (Reference Bhalotra, Brulé and Roy2020)].Footnote 1 When son preference, as shown in the literature, leads to discriminatory behavior against the female fetus, the birth outcomes of newborn girls are not likely to be as good as those of boys, because prenatal care has a significant impact on the birth outcomes of newborns [Gajate-Garrido (Reference Gajate-Garrido2013), Sonchak (Reference Sonchak2015)]. However, only a few studies have examined this relationship.
We analyzed the relationship between son preference and birth weight in girls.Footnote 2 The data cover the period between 2000 and 2015 from South Korea, which has a long history of strong son preference. Recently, however, a decline in son preference has become apparent. The proportion of married women who answered that they wanted to have a son was 18.0% in 2000 but fell to 5.7% in 2015 [Oh et al. (Reference Oh, Kim, Shin and Bae2016)]. In addition, married women who were not concerned about not having a son accounted for 38.2% in 2000, but this number increased to 65.3% in 2015 [Oh et al. (Reference Oh, Kim, Shin and Bae2016)]. The persistence of son preference, followed by its weakening is rarely observed; thus, the South Korean situation is appropriate for analysis.Footnote 3
The degree of son preference is measured by the sex ratio at birth (SRB), which is the number of newborn boys per 100 newborn girls. Since almost all pregnant women underwent ultrasound tests in South Korea before 2000 and abortion has been allowed in practice, the SRB can reflect son preference. Accordingly, other studies have used SRB as a measure of South Korea's son preference [e.g., Chung and Gupta (Reference Chung and Gupta2007), Den Boer and Hudson (Reference Den Boer and Hudson2017), Choi and Hwang (Reference Choi and Hwang2020)]. If the SRB reflects son preference, a stronger son preference translates to a larger SRB. In other words, parents with a son preference who are likely to follow the son-stopping rule (continuing childbearing until they have the desired number of sons) when they do not have access to ultrasound technology and abortion could tend to abort the female fetus when they do, which leads to a larger SRB.
As shown in Figure 1, the SRB in 2000 for first- or second-born children was in the normal range at 106.3 and 107.4,Footnote 4 respectively; whereas, the SRB of third- or later-born children in the early 2000s, which was as high as 144.2 in 2000 [National Statistical Office (2020)], was not. Therefore, we focus our analysis on the children of parents that were considered to have a son preference, that is, third- or later-born children, and first- and second-born children are used for the placebo tests.
Figure 1. SRB based on birth order.
This figure shows SRB over time based on birth order.
SRB is also affected by other factors. For example, malnutrition is a factor that increases the rate of girls being born [Anderson and Bergström (Reference Anderson and Bergström1998), Almond et al. (Reference Almond, Edlund, Li and Zhang2010)], and natural disasters such as earthquakes increase the proportion of girls [Fukuda et al. (Reference Fukuda, Fukuda, Shimizu and Møller1998)].Footnote 5 However, malnutrition and earthquakes are not considered serious problems in South Korea. Furthermore, based on the Trivers-Willard hypothesis, Korean women's increased education level and age at birth over time, may have decreased and increased the proportion of girls, respectively. However, as Figure 1 shows, the sex ratio for first- and second-born children is stable over time, indicating that these variables are not likely to have affected the ratio in the country.
Our study relates to the literature that analyzes discriminatory behavior against girls arising from son preference. In addition to studies on prenatal discrimination in prenatal hospital visits and abortions, postnatal discrimination has also been reported. These studies have shown that girls are discriminated against for immunizations, breastfeeding, mortality, childcare time, child labor, and education [Oster (Reference Oster2009), Jayachandran and Kuziemko (Reference Jayachandran and Kuziemko2011), Kumar (Reference Kumar2013), Barcellos et al. (Reference Barcellos, Carvalho and Lleras-Muney2014), Guilmoto (Reference Guilmoto2015), Hafeez and Quintana-Domeque (Reference Hafeez and Quintana-Domeque2018), Kaul (Reference Kaul2018), Choi and Hwang (Reference Choi and Hwang2020)]. For example, Choi and Hwang (Reference Choi and Hwang2020) reported that in South Korea, boys spend less time on household chores, while their mothers work fewer hours in labor markets. Studies on the effect of discriminatory behavior against girls on birth outcomes are rare. A related study was conducted by Lhila and Simon (Reference Lhila and Simon2008), who analyzed the data of first-generation Chinese and Indian mothers living in the United States and found that knowing the gender of the fetus had no significant effect on girls' weight at birth.Footnote 6
Our study also relates to the literature on factors associated with birth weight. These factors include the gender and birth order of the child [Goisis et al. (Reference Goisis, Remes, Martikainen, Klemetti and Myrskylä2019)] and maternal characteristics such as age [Restrepo-Méndez et al. (Reference Restrepo-Méndez, Lawlor, Horta, Matijasevich, Santos, Menezes, Barros and Victora2014)], income [Conley and Bennett (Reference Conley and Bennett2001)], and educational attainment [Ahsan and Maharaj (Reference Ahsan and Maharaj2018)]. These factors also include maternal stress during pregnancy, possibly from natural disasters or economic conditions [Currie and Rossin-Slater (Reference Currie and Rossin-Slater2013), Bozzoli and Quintana-Domeque (Reference Bozzoli and Quintana-Domeque2014), de Oliveira et al. (Reference de Oliveira, Lee and Quintana-Domequeforthcoming)] and health behaviors such as prenatal care [Gajate-Garrido (Reference Gajate-Garrido2013), Sonchak (Reference Sonchak2015)].
This study contributes to the literature in the following ways. First, as described above, many studies have analyzed discriminatory behavior against girls, but very few have analyzed its effect on birth outcomes. As birth outcomes affect later health and school outcomes [Figlio et al. (Reference Figlio, Guryan, Karbownik and Roth2014), Bharadwaj et al. (Reference Bharadwaj, Eberhard and Neilson2018), McGovern (Reference McGovern2019)], they represent an essential factor in determining the level of human capital. For example, Bharadwaj et al. (Reference Bharadwaj, Eberhard and Neilson2018) reported that low birth weight (that is, weighing less than 2,500 grams at birth; LBW) decreases math test scores by 0.1 standard deviations, and McGovern (Reference McGovern2019) found that LBW is associated with a six percentage point increase in mortality risk.Footnote 7 Second, we used SRB to measure the degree of son preference, which is possible because the rate has declined in South Korea with the availability of abortion and ultrasound technology throughout the country. China and India have seen a recent rise; however, this increase should be interpreted as a rise in the prevalence of ultrasound technology [Ebenstein (Reference Ebenstein2010), Bhalotra and Cochrane (Reference Bhalotra and Cochrane2010), Hu and Schlosser (Reference Hu and Schlosser2015)], not as a change in son preference. Instead, this ratio has been used for other purposes. For example, Hu and Schlosser (Reference Hu and Schlosser2015) examined the effect of sex-selective abortion in India on girls' well-being and used state-level SRB to measure how common abortions are in each state. The use of state-level SRB was possible because of significant differences in prenatal sex selection technology over time and across states in the country, which implies that not every state had the technology during the period examined in the study: the early 1980s to the mid-2000s. In this context, although Jayachandran and Kuziemko (Reference Jayachandran and Kuziemko2011) used SRB to measure son preference in Indian states, Hu and Schlosser (Reference Hu and Schlosser2015) caution against the use of the ratio to measure son preference in India, as their findings indicate that there were Indian states with strong son preference but with no trend of increases in SRB.
The results of this study are as follows. Girls are more likely to be born with LBW when son preference is stronger. Specifically, if the ratio increases by one, the probability of LBW among girls increases by 0.0156 percentage points. Because SRB for third- or later-born children decreased by 38.6 from 144.2 in 2000 to 105.6 in 2015, the decline in SRB was associated with a reduction in the probability of LBW among girls by 0.6 percentage points (= 0.0156%p × 38.6). In addition, when the ratio is higher, girls are more likely to be born outside hospitals, implying that mothers conceiving girls make fewer prenatal visits to the hospital when the son preference is stronger.
The remainder of this paper is organized as follows. Section 2 describes the background, followed by description of the data in Section 3. Section 4 describes the empirical strategy. Section 5 presents the estimation results, and Section 6 concludes the paper.
2. Background
2.1 Son preference in South Korea
South Korea, along with China and India, has been a country with strong son preference; this is rooted in the patriarchal family system introduced during the Choson dynasty, which reigned from 1392 to 1910. Under their patriarchal family system, the son (or the eldest son) becomes head of the family, and inheritance occurs through the male line, leading to a strong son-preferring culture [Larsen et al. (Reference Larsen, Chung and Gupta1998), Chung and Gupta (Reference Chung and Gupta2007)].
Son preference has declined and almost disappeared in recent times. As described in Section 1, in 2015 only 5.7% of married women thought that they wanted to have a son, compared to 18.0% in 2000. Furthermore, according to the Korean General Social Survey (KGSS), in 2004, more people, in the case of having only one child, preferred to have a son rather than a daughter, accounting for 36.2% and 31.9%, respectively. However, in 2015, 43.1% preferred having a daughter, whereas 35.6% preferred to have a son.Footnote 8
Figure 2 shows the SRB for third- or later-born children during 2000–2015 for 16 regions in South Korea; there is a decline in the ratio in all the regions.Footnote 9 Larger declines occurred in the southeastern region, which has a strong tradition of son preference, including Busan, Daegu, and Ulsan. For example, Daegu had the largest decline, from 192.1 in 2000 to 106.2 in 2015; the ratio in Seoul, which had a relatively weak son preference, dropped from 137.9 to 104.2 during the same period.
Figure 2. Regional SRB of third- or later-born children.
These graphs are the ratios of third- or later-born children for 16 regions in South Korea.
The improved socioeconomic status of women is one reason why son preference has declined in South Korea [Chung and Gupta (Reference Chung and Gupta2007), Edlund and Lee (Reference Edlund and Lee2013)]. Women's participation in economic activities has increased, as has their educational attainment. Girls perform better than boys on standardized tests and are more likely to attend college. As reported in the literature, discrimination against women decreases when their economic conditions improve. According to Qian (Reference Qian2008), when the prices of female-grown agricultural products and women's income rise, girls are less likely to die and women's educational level increases. Jensen (Reference Jensen2012) showed that as the number of female-labor intensive jobs increases, girls receive more education.
2.2 Abortion and ultrasound technology
Although abortion is illegal in South Korea except for in certain cases, it is widely performed. In 2005, there were approximately 340,000 abortions and in 2010, 170,000 [Sohn et al. (Reference Sohn, Kang, Chang, Kim, Park, Nam and Kang2011)].Footnote 10 Nevertheless, only few people have been indicted; for example, in 2006, only five cases led to the indictment [Korean Women's Association United (2011)]. This means that abortion has been allowed in practice in the country.
According to Hong et al. (Reference Hong, Lee, Chang, Oh and Kye1994), 97.6% of pregnant women underwent an ultrasound test in 1994, which implies that a pregnancy ultrasound test was available in every region of the country in the years prior to 2000, and the abnormally high SRB for third- or later-born children was caused by sex-selective abortion of girls enabled by the test.
It is worth noting that as abortion is allowed in practice, the estimates found in this study are likely to be smaller than those estimated when abortion is not allowed. We consider that mothers who tend to abort the female fetus when abortion is allowed, are more likely to treat the baby improperly (e.g., abandonment) when abortion is not allowed than mothers who do not abort and carry the pregnancy of a girl to term when it is allowed. Thus, girls who would have been born to mothers who intend to carry out an abortion when it is prohibited, are likely to have worse birth outcomes than girls who are born when it is allowed. This implies that gender differences in birth outcomes would have been larger when abortion was not allowed.
2.3 Fertility
Although other son-preferring countries have seen an increase in SRB as fertility declined [Ebenstein (Reference Ebenstein2010), Jayachandran (Reference Jayachandran2017)], South Korea has seen declines in both SRB and fertility simultaneously: it currently has one of the lowest fertility rates in the world. The total fertility rate was 2.8 in 1980 but declined to 1.5 in 2000 and to 1.2 in 2015 [National Statistical Office (2017)]. This low fertility rate, combined with the abnormally high SRB for third- or later-born children, indicates that son preference is limited to parents with three or more children.
3. Data
Data from 2000 to 2015 were obtained from the birth registry maintained by the National Statistical Office. The data contain information on each baby's gender, birth year/month, region, birth weight, birth order, and birthplace (e.g., hospital, home, etc.). The data also contain parental information including age and educational attainment. The birth regions comprise the seven largest cities, including Seoul and nine provinces consisting of several small or mid-sized cities.
The analysis sample was approximately 760,000 third- or later-born children born between 2000 and 2015. Approximately 66,000 of them were born in 2000, the year with the highest number of births, and approximately 43,000 were born in 2015, which recorded the lowest number of births. Gyeonggi, a province surrounding Seoul, had the largest number of babies (approximately 180,000), followed by Seoul, which had approximately 110,000 births (median = 38,000). Table 1 presents the descriptive statistics. In column (1) for third- or later-born children, girls accounted for 45.2% of the total, and 5.5% had LBW. In addition, the proportion of mothers with college degrees was 41.6%. This number increased from 23.3% in 2000 to 62.7% in 2015, reflecting the gradual socioeconomic improvements in women's lives. Columns (2) and (3) of the table indicate that girls account relatively more for other birth orders, that is, 48.7% and 48.6% for first- and second-born children, respectively.
Table 1. Descriptive statistics
Standard deviations are in parentheses. Low birth weight means weighing less than 2,500 g at birth.
4. Empirical strategy
We estimated the following regression equation using ordinary least squares and examined whether son preference affects birth outcomes for girls. The subscripts i, r, and t represent newborn baby, region, and year, respectively.
The dependent variable includes birth weight and a dummy variable indicating LBW. The independent variables include the variable girl, a dummy variable indicating whether a baby is a girl, and SRB representing SRB at the region/year level. Girl × SRB is the interaction of the two variables, and the coefficient for this variable shows how birth outcomes for girls change relative to birth outcomes for boys as SRB changes. The coefficient sign is expected to be negative when the dependent variable is birth weight and son preference leads to less prenatal care for girls. If the dependent variable is a dummy variable that indicates LBW, it is expected to be positive. Vector X includes the mother's age, its square, and a dummy variable indicating whether the mother is a college graduate. The vector also includes the interactions of these variables with the variable girl. Vector I includes regional income, namely, individual income per capitaFootnote 11 and its interaction with the variable girl. In the equation, γ, δ, and ɛ are the region, year fixed effect, and error term, respectively. Finally, standard errors are clustered at the region/year level. We also calculated standard errors clustering at the region level using the method suggested by Cameron et al. (Reference Cameron, Gelbach and Miller2008), but the results did not change.
Among the factors for birth weight described in Section 1, we did not control for prenatal care and maternal stress. Prenatal care should not be controlled because SRB is likely to affect prenatal care and controlling for it could lead to an over-control problem. Not controlling for maternal stress is problematic when it affects SRB and birth outcomes differently based on gender. As a robustness check, we controlled for the interaction of region and year fixed effects to check whether stress-inducing events possibly occurring in a particular region/year confounds the estimate.Footnote 12
5. Results
5.1 The effect of son preference on girls' birth outcomes
Table 2 presents the estimation results for equation (1). The coefficients and their standard errors in column (2) were multiplied by 100. Therefore, if a coefficient is 0.02, it can be interpreted as 0.02 percentage points rather than two percentage points. Focusing on LBW in column (2), the coefficient of the interaction between SRB and the dummy variable girl is 0.0156, meaning that if the ratio increases by one, the probability of LBW among girls increases by 0.0156 percentage points relative to the probability for boys.Footnote 13 As the SRB for third- or later-born children decreased from 144.2 in 2000 to 105.6 in 2015, with a decrease of 38.6, the decline in SRB was related to a reduction in the probability of LBW among girls by 0.60 percentage points ( = 0.0156%p × 38.6). Furthermore, because the ratio of newborns with LBW was 5.6% (Table 1), 0.60 percentage points was equivalent to 10.7% (= 0.0060/0.056). In other words, during 2000–2015, a decrease in SRB was related to a 10.7% reduction in the probability of LBW among girls. Finally, one can see in Table 2 that a one-unit increase in SRB is related to a decline in the probability of LBW among boys by 0.0079 percentage points and an increase in the probability among girls by 0.0077 (= − 0.0079 + 0.0156) percentage points, implying that the reduced probability of LBW among girls by 10.7% over the period is related to an increase in the probability among boys and a decrease in the probability among girls.
Table 2. Effect of son preference on girls' birth outcomes
Standard errors are in parentheses. They are clustered at the region/year level. These regressions also include a constant, two dummy variables indicating a girl and whether a mother has a college degree, maternal age, age squared, regional income, their interactions with a girl, region fixed effect, and year fixed effect. The coefficients and their standard errors in column (2) are multiplied by 100.
*: p < 0.05.
5.2 Results for other children (placebo tests)
We conducted the same analysis reported in Table 2 for first- and second-born children as placebo tests. Considering that son preference did not exist for these children, their birth outcomes should be unaffected. In Table 3, columns (1) and (2) are for first-born children, and columns (3) and (4) are for second-born children. As the table shows, no coefficients for girl × SRB were statistically significant, reflecting that the parents of first- and second-born children had no son preference.
Table 3. Results for other children (placebo tests)
Standard errors are in parentheses. They are clustered at the region/year level. Other independent variables are the same as in Table 2. The coefficients and their standard errors in columns (2) and (4) are multiplied by 100.
5.3 Mechanism of the effect
The literature shows that son preference leads to less prenatal care for girls. To examine whether less prenatal care or less frequent hospital visits are related to the results in Table 2, we examined whether son preference is related to an increase in the incidence of non-hospital births for girls. This is because non-hospital births are likely to reflect fewer hospital visits. For the analysis, we estimated regression equation (1) with a dummy dependent variable indicating whether a baby was born outside a hospital. The results are presented in column (1) of Table 4, and the coefficients and their standard errors are multiplied by 100. This column shows that if the ratio increases by one, the probability of a girl being born outside the hospital increases by 0.0075 percentage points relative to the probability for boys. As the SRB of third- or later-born children decreased by 38.6 during the period 2000 to 2015, the decline in the son preference was related to a reduction in the probability of a girl being born outside a hospital by 0.29 percentage points ( = 0.0075%p × 38.6), which is equivalent to 13.2% ( = 0.0029/0.022) because the proportion of non-hospital-born children during 2000–2015 is 2.2%, as shown in Table 1. This result implies that mothers with female fetuses are likely to make fewer prenatal hospital visits when the son preference is stronger, which could be an important reason why son preference leads to low birth weight for girls. Finally, we conducted the same analysis on first- and second-born children. As presented in columns (2) and (3) of Table 4, the coefficient of the interaction term is insignificant, indicating that girls of these birth orders did not go through prenatal discrimination.
Table 4. Mechanism analysis
Standard errors are in parentheses. They are clustered at the region/year level. Other independent variables are the same as in Table 2. The coefficients and their standard errors are multiplied by 100.
*: p < 0.05.
5.4 Robustness checks
We conducted two robustness checks for the results presented in Table 2. First, we controlled for the interaction of region and year fixed effects to check whether an event occurring in a particular region/year confounds the estimate. It is noteworthy that when conducting this analysis, SRB and regional income were dropped because these variables were at the region/year level. As shown in panel A of Table 5, the inclusion of the interaction changes the result little. The coefficient for girl × SRB in column (2) is 0.0162, while the corresponding coefficient in Table 2 is 0.0156. Second, we used two- and three-year moving averages for SRB instead of the current year SRB, because babies born at the beginning of a year are conceived a year before and are not likely to be affected by the son preference of the year in which they were born. Panels B and C of Table 5 indicate that the results do not differ from those in Table 2. In summary, the results in Table 2 are robust to the different specifications. Finally, we conducted the same analysis reported in Table 2, excluding southeastern regions with a strong son preference (one region at a time), to examine whether a particular region drove the result in Table 2. We conducted the same analysis excluding Gyeonggi and Seoul, the two largest regions. In Table 6, focusing on LBW in column (2), all estimates are significant, ranging from 0.0110 to 0.0176, implying that the results in Table 2 are robust to the sample.
Table 5. Robustness checks
Standard errors are in parentheses. They are clustered at the region/year level. Other independent variables are the same as in Table 2, although the analysis in panel A does not control for SRB and regional income because these variables are at the region/year level. The coefficients and their standard errors in columns (2) and (3) are multiplied by 100.
*: p < 0.05.
Table 6. Effect of son preference on girls' birth outcomes and birthplace excluding one region at a time
Standard errors are in parentheses. They are clustered at the region/year level. Other independent variables are the same as in Table 2. Panel A excludes Busan, and so on. The coefficients and their standard errors in columns (2) and (3) are multiplied by 100.
*: p < 0.05.
5.5 Heterogeneity analysis
This subsection examines the effects based on the period. As presented in Figure 1, the SRB of third- or later-born children decreased faster in the period 2000–2007 than in 2008–2015, which implies that the effect is likely to be more significant in the former period than in the latter. Table 7 shows that the relationship is significant only between 2000 and 2007. The coefficient for the interaction term is −0.2447 in column (1) and 0.0174 in column (2).
Table 7. Analysis based on the period
Standard errors are in parentheses. They are clustered at the region/year level. Other independent variables are the same as in Table 2. The coefficients and their standard errors in columns (2) and (4) are multiplied by 100.
*: p < 0.05.
6. Conclusions
Research has shown that there is discrimination against the female fetus when the son preference is strong. Girls of parents with strong son preference are less likely to receive prenatal care and more likely to be aborted. As prenatal care is an important factor in determining a baby's birth outcomes, those for girls are likely to be worse when the son preference is strong. However, only a few studies have examined this topic.
To analyze this topic, we measured the degree of son preference using SRB, which was possible because the rate declined in South Korea with the availability of abortion and ultrasound technology throughout the entire country. We found that girls were more likely to be born with LBW when son preference was stronger. In addition, we present evidence that girls are more likely to be born outside hospitals in these cases, which implies that girls are discriminated against prenatally. In summary, over the period 2000–2015, the decline in SRB for third- or later-born children by 38.6 reduced the probability of a girl being born outside hospitals by 13.2%, while the probability of LBW for them decreased by 10.7%.
Supplementary material
The supplementary material for this article can be found at https://doi.org/10.1017/dem.2022.13.
Acknowledgements
The author is grateful to Bo Min Kim, Jihyun Kwon, Tae Hee Kwon, Yong Woo Lee, Chuhwan Park, and seminar participants at the Asian and Australasian Society of Labour Economics, Korean Economic and Business Association, Korean Applied Economic Association, and Western Economic Association International Conference for their valuable comments on this research. This research was supported by the 2021 Yeungnam University research grant.
