Learning to Value Girls: Balanced Infant Sex Ratios at Higher Parental Education in the United States, 1969–2018

Infant sex ratios that differ from the biological norm provide a measure of gender status inequality that is not susceptible to social desirability bias. Ratios may become less biased with educational expansion through reduced preference for male children. Alternatively, bias could increase with education through more access to sex-selective medical technologies. Using National Vital Statistics data on the population of live births in the United States for 1969–2018, we examine trends in infant sex ratios by parental race/ethnicity, education, and birth parity over five decades. We find son-biased infant sex ratios among Chinese and Asian Indian births that have persisted in recent years, and regressions suggest son-biased ratios among births to Filipino and Japanese mothers with less than a high school education. Infant sex ratios are more balanced at higher levels of maternal education, particularly when both parents are college educated. Results suggest greater equality of gender status with higher education in the United States.


Records
. Values indicate the proportion of birth records missing information for each variable. The first column includes all years . The other columns include 1969-1994 and 2011-2018, years with paternal education information. The total number of birth records in all years is higher than the number used in this paper because they include birth orders higher than 3 and racial/ethnic groups not examined here (e.g., Guamanian, Samoan). Records with paternal education have non-missing paternal education.  1969-19741970Census 1975-19841980Census 1985-19941990Census 1995-20002000Census 2001-20052001-2005 NVSS 1969NVSS -2018 to births at parities 1-3 (live birth orders 1-3). All models include indicators for each race/ethnic category and year. Full model includes controls for mean maternal and paternal age, marital status, proportion of mothers who live in the U.S., and aggregate group characteristics from Census and ACS data measured separately by gender: proportions not in the labor force, foreign born, living in a three-generation household, non-U.S. citizen, home ownership, Hispanic ethnicity, farm residence, and mean values of years living in the U.S., family income, and home values. Bootstrapped standard errors stratified by race/ethnicity in parentheses. ** p<0.01, * p<0.05, + p<0.1    NVSS 1969NVSS -2018 to births at parities 1-3 (live birth orders 1-3). All models include indicators for each race/ethnic category, birth order, and year. Full model includes controls for mean maternal and paternal age, marital status, proportion of mothers who live in the U.S., proportion of births at 1 st , 2 nd , and 3 rd parity, and aggregate group characteristics from Census and ACS data measured separately by gender: proportions not in the labor force, foreign born, living in a three-generation household, non-U.S. citizen, home ownership, Hispanic ethnicity, farm residence, and mean values of years living in the U.S., family income, and home values. Bootstrapped standard errors stratified by race/ethnicity in parentheses. ** p<0.01, * p<0.05, + p<0.1

Supplementary Online Appendix: Analyses Conditional on Sex of Previous Children
Data and Methods Using pooled American Community Survey data for years 2000-2018 from IPUMS (Ruggles et al. 2018), we follow previous research (Almond and Edlund 2008) and restrict the sample to families in which: 1) one parent is the household head; 2) all children were born in the U.S.; 3) none of the children are adopted or step-children; and 4) the eldest child is less than 13 years old to reduce the chances that an older child has left the household. We focus on the eldest three children (parity one through three) and further exclude children who are from multiple births (e.g., twins). This yields an analysis sample of 2,945,642 children in 971,150 families.
Within this sample, we assign parity based on the age of children in the household and examine child sex ratios by parity and the sex of older children. We focus particularly on 175,311 children born at third parity with either two previous girl or two previous boy children.
Parental education categories include those with less than high school, a high school degree, some college, or at least a college degree. We estimate the likelihood of having a boy at third parity among parents with no college education compared to those with at least some college education. We examine variation between parents with and without any college because it enhances income potential (Averett and Burton 1996;Giani et al. 2020) and exposes people to more cosmopolitan cultural beliefs (Baker 2014;Greenwood 1975). We examine variation by maternal education and we also compare by paternal education because male attitudes, preferences, and earning potential hold more cultural value in traditional or patriarchal cultures (e.g., China, South Korea, and India;Jha et al. 2006;Yi et al. 1993;Chung and Gupta 2007;Gietel-Basten et al. 2018). In sensitivity analyses, we find similar results when using paternal identification (rather than maternal) to define race/ethnicity category and find similar results.
For each third child with either two older brothers or two older sisters, we estimate the likelihood of being a boy. Out of concern about models that use non-linear link functions (Mood 2010; Gomila 2020), we use linear probability models to predict likelihood of a boy child at third parity. Equation A1 predicts the likelihood that a third child (i) is a boy, with the sex mix of previous children, and indicators for each maternal race/ethnic category (j), education category (k), and year (t). Robust standard errors are adjusted for clustering within maternal race/ethnic category. The coefficients of interest ( ) test whether the likelihood a third child with two sisters is a boy differs by race/ethnic category. For each race/ethnic group, these coefficients compare the likelihood of having a boy at third parity when the two previous children are girls rather than boys. These analyses allow a further test of hypotheses 1a and 1b in the main analyses, conditional on the sex of previous children.

Boy
= Sisters *Race/Ethnic + Educ + Year + + To test the other hypotheses, we limit the sample to third children with two older sisters. Using this sample, coefficients for maternal education categories ( ) interacted with each race/ethnic category test whether the likelihood of having a boy after two previous girls differs significantly by maternal education (hypothesis 2). Coefficients for a continuous year measure interacted with race/ethnic category test whether trends differ in likelihood of having a boy after two previous girls (hypothesis 3). To test hypothesis 4, we stratify the sample by paternal college education and repeat analyses for hypothesis 2, with interaction terms for maternal education and race/ethnic category. We follow the same approach as the main analyses to test whether coefficients for maternal college education by race/ethnicity differ significantly among children of fathers with and without college education (Paternoster et al. 1998;Clogg et al. 1995).
We fit these models with and without controls for parental and family characteristics (W). Parental characteristics include maternal and paternal age and marital status. Economic measures include separate indicators for whether the mother or father is not in the labor force, indicators for home ownership and residence on a farm, home value (logged to reduce skewness), and family and household income (logged). Cultural measures include separate indicators for whether the mother or father is foreign born or a non-citizen, separate measures for years the mother and father have resided in the U.S. (among those born outside the country), and an indicator for living in a three-generation household (e.g., grandparents, parents, and children). The indicator for a mother who is not in the labor force is included in the economic measures; we also include it in the cultural measures because it provides a proxy for traditional gender roles.
We fit models with and without these controls because previous work suggests that child sex ratios may depend on these economic and cultural characteristics (Guilmoto 2009;Bongaarts 2013). An association between parental education and child sex could be confounded by economic or cultural measures without holding them constant. Table B2 shows predicted likelihood of having a male third child by maternal race/ethnicity and sex of the two previous children. Figure B1 shows predictive margins based on the full model in Table B2. Among white mothers, third children with two previous girls are slightly, but significantly less likely to be male. This is consistent with existing evidence of a tendency to repeat the sex of earlier children (Almond and Edlund 2008;Renkonen et al. 1962). However, estimates vary by maternal race/ethnicity and suggest a significantly higher likelihood of having a boy among Chinese, Asian Indian, Korean, Japanese, and Filipino mothers when the two previous children are girls. Figure B1 shows that the difference is particularly pronounced among children of Chinese, Asian Indian, and Japanese mothers, who are respectively 4.6%, 12.7%, and 4.8% more likely to be male, when they have two older sisters compared to children of white mothers. Consistent with hypothesis 1, these estimates suggest male-biased child sex ratios among Chinese, Asian Indian, and Japanese families, when the two previous children are girls.

Results
If parents from these groups are using sex selective behaviors to increase the likelihood of having a boy, then there should be a longer gap between the second and third children for those groups. Results support this; Table B3 shows estimates predicting the age gap between second and third children when the two previous children are girls. Compared to children of white mothers, Figure B2 shows there is a significantly longer predicted gap between second and third children among Chinese, Asian Indian, and Vietnamese mothers when the two previous children are girls. Among Chinese and Asian Indian mothers, this amounts to approximately 2-3 months longer between children. Table B4 examines trends in the likelihood of having a boy when the two previous children are girls. There is no trend among whites and for most groups there is a slight increase over time in the likelihood of a male third child after two previous girls. However, among Asian Indian and Filipino mothers, there is a slight, but significant decline of about 0.5% per year since 2000. This offers some support for hypothesis 3 regarding gender inequality, but only suggests increasing equality among Asian Indian and Filipino parents. Table B5 shows results of models testing for variation by maternal education. For nearly every group, third children with two older sisters are significantly less likely to be boys when the mother has attended college. In the full model, these estimates range from 1% lower likelihood of a boy among Chinese to 14% among Japanese when the mother has more education. These results support hypothesis 2. The exceptions are third children of Asian Indian and African American mothers, who are more likely to be a boy when the mother has attended college (4% and 3%, respectively). This could reflect less exposure to environmental toxins among those with higher education, which is related to lower child sex ratios (Sakamoto et al. 2001;Figa-Talamanca and Petrelli 2000;Whorton et al. 1994;Goldsmith et al. 1984). Table B6 compares the relationship between maternal education and likelihood of a male third child by paternal education. Coefficients for maternal college education differ significantly between models by paternal education for all race/ethnic groups (Paternoster et al. 1998). In the full model, maternal college education is related to lower likelihood of having a boy when the father has also attended college for all groups except African Americans and Vietnamese. Estimates suggest maternal education is related to a higher likelihood of having a boy for several groups when the father has not attended college. Specifically, compared to white mothers, Chinese and Asian Indian mothers are over 10% more likely to have a boy after two girls when they have attended college and the father has not. However, this relationship reverses when both parents have attended college and Chinese and Asian Indian mothers are about 10% less likely to have a boy after two girls. Results for Korean mothers also suggest maternal college education is related to more equal infant sex ratios when both parents have attended college. The positive relationship between parental college education and likelihood of having a male third child among African Americans and Vietnamese are consistent with greater equality of child sex ratios because of the female-biased ratios among African Americans (see Figure 1) and the significantly lower likelihood of having a male third child among Vietnamese (see Table B2). Overall, results support hypothesis 4 and suggest greater equality of child sex ratios with maternal education when parents are educationally homogamous.
Results are similar if we include families with likely multiple births (i.e. children of equal age and parity). Results are also consistent when excluding families in which the mother or father is not a U.S. citizen. Overall, results examining child sex ratios conditional on the sex of previous children are consistent with the main analyses. Results do not support the argument that education increases son-biased sex ratios by increasing access to sex-selective technologies. Rather, results indicate that education is related to less male-biased preferences among Asian parents. Education is more strongly related to equality of child sex ratios when both parents have attended college, which is consistent with the suggestion that higher education promotes more egalitarian views. ACS 2000-2018, limited to children 1-3 in households where: one parent is the household head; all children were born in the U.S.; no children are adopted or step-children; the eldest child is less than 13 years old; and children are not from multiple births.  -0.008** -0.007** -0.007** -0.007** -0.007** (0.000) (0.000) (0.000) (0.000) (0.000) African American * Two Previous Girls 0.004** 0.007** 0.006** 0.007** 0.007** (0.000) (0.000) (0.000) (0.000) (0.000) American Indian * Two Previous Girls 0.006** 0.010** 0.010** 0.010** 0.010** (0.000) (0.000) (0.000) (0.000) (0.000) Chinese * Two Previous Girls 0.045** 0.044** 0.045** 0.045** 0.046** (0.000) (0.000) (0.000) (0.000) (0.000) Asian Indian * Two Previous Girls 0.119** 0.127** 0.128** 0.126** 0.127** (0.000) (0.000) (0.000) (0.000) (0.000) Korean * Two Previous Girls 0.024** 0.024** 0.024** 0.024** 0.024** (0.000) (0.000) (0.000) (0.000) (0.000) Japanese * Two Previous Girls 0.042** 0.047** 0.047** 0.047** 0.048** (0.000) (0.000) (0.001) (0.000) (0.001) Filipino * Two Previous Girls 0.013** 0.018** 0.018** 0.019** 0.019** (0.000) (0.000) (0.000) (0.000) (0.000) Vietnamese * Two Previous Girls 0.000* -0.005** -0.005** -0.005** -0.006** (0  ACS 2000ACS -2018 to third children with two older brothers or sisters in households where: one parent is the household head; all children were born in the U.S.; no children are adopted or step-children; the eldest child is less than 13 years old; and children are not from multiple births. All models include indicators for each race/ethnic category, maternal education category, and year. Full model includes controls for parental characteristics :maternal and paternal age, parental marital status; economic characteristics: separate indicators for whether the mother or father is not in the labor force, indicators for home ownership and residence on a farm, home value (logged), and family and household income (logged); and cultural measures: indicator for whether mother is not in labor force, separate indicators for whether the mother or father is foreign born or a non-citizen, separate measures for years the mother and father have resided in the U.S., indicator for living in a three-generation household (e.g., grandparents, parents, and children). Robust standard errors adjusted for race/ethnic clustering in parentheses. ** p<0.01, * p<0.05, + p<0.1  ACS 2000ACS -2018 to third children with two older sisters in households where: one parent is the household head; all children were born in the U.S.; no children are adopted or step-children; the eldest child is less than 13 years old; and children are not from multiple births. All models include indicators for each race/ethnic category, maternal education category, and year. Full model includes controls for parental characteristics :maternal and paternal age, parental marital status; economic characteristics: separate indicators for whether the mother or father is not in the labor force, indicators for home ownership and residence on a farm, home value (logged), and family and household income (logged); and cultural measures: indicator for whether mother is not in labor force, separate indicators for whether the mother or father is foreign born or a non-citizen, separate measures for years the mother and father have resided in the U.S., indicator for living in a three-generation household (e.g., grandparents, parents, and children). Robust standard errors adjusted for race/ethnic clustering in parentheses. ** p<0.01, * p<0.05, + p<0.1   ACS 2000ACS -2018 to third children with two older sisters in households where: one parent is the household head; all children were born in the U.S.; no children are adopted or step-children; the eldest child is less than 13 years old; and children are not from multiple births. All models include indicators for each race/ethnic category, maternal education category, and year. Full model includes controls for parental characteristics :maternal and paternal age, parental marital status; economic characteristics: separate indicators for whether the mother or father is not in the labor force, indicators for home ownership and residence on a farm, home value (logged), and family and household income (logged); and cultural measures: indicator for whether mother is not in labor force, separate indicators for whether the mother or father is foreign born or a non-citizen, separate measures for years the mother and father have resided in the U.S., indicator for living in a three-generation household (e.g., grandparents, parents, and children). Robust standard errors adjusted for race/ethnic clustering in parentheses. ** p<0.01, * p<0.05, + p<0.1