Dating a mixed guy when your white

Objectives. I sought to identify whether there were associations between paternal race/ethnicity and birth outcomes among infants with parents of same- and mixed-races/ethnicities.

Methods. Using the National Center for Health Statistics 2001 linked birth and infant death file, I compared birth outcomes of infants of White mothers and fathers of different races/ethnicities by matching and weighting racial/ethnic groups following a propensity scoring approach so other characteristics were distributed identically. I applied the same analysis to infants of Black parents and infants with a Black mother and White father.

Results. Variation in risk factors and outcomes was found in infants of White mothers by paternal race/ethnicity. After propensity score weighting, the disparities in outcomes by paternal or parental race/ethnicity could be largely attributed to nonracial parental characteristics. Infants whose paternal race/ethnicity was unreported on their birth certificates had the worst outcomes.

Conclusions. The use of maternal race/ethnicity to refer to infant race/ethnicity in research is problematic. The effects of maternal race/ethnicity on birth outcomes are estimated to be much larger than that of paternal race/ethnicity after I controlled for all covariates. Not listing a father on the birth certificate had a strong association with outcomes, which might be a source of bias in existing data and a marker for identifying infants at risk.

Data

The NCHS 2001 birth cohort data consists of all live births and all deaths from 2001 linked to the corresponding birth certificates. This data set has rich information on birth outcomes (e.g., birthweight, 5-minute Apgar score) and parents’ demographic and socioeconomic background (e.g., race/ethnicity, maternal education).

The analyzed data set was reduced in several ways. I excluded non-White mothers and twins from the main analysis. The original sample size (i.e., the birth cohort of 2001) was 4 025 933 live births, among which 128 717 (3%) were twins or plural births. (Table 1 provides an overview of infants of mixed-race and same-race parents and offers some sense of why I chose to focus on White mothers only: both Black mothers and Mexican mothers have a high rate of male partners unreported on the birth certificate.) California data were excluded from these analyses because maternal tobacco and alcohol use during pregnancy are important behavioral risk factors that were included as variables in this study but these were not reported on California birth certificates. Because I focused on both maternal and paternal influence on birth outcomes and there is a high proportion of unreported fathers for Black and Mexican mothers, the primary analysis was of the singleton live births to White mothers. Comparisons with White and Black mothers and their partners are briefly discussed at the end of the “Results” section. Seven paternal race/ethnicity categories were selected, each comprising at least 0.15% of the subpopulation of White mothers. After restricting the data according to these criteria, 2 054 542 births in 2001 were available for analysis.

Measures

Independent variables that were considered to be potential risk factors for birth outcomes based on previous studies included parental characteristics and behaviors and children's characteristics. Previous research has shown nonlinear effects of parental age on infant outcomes¹⁷; therefore, both maternal and paternal age were included and categorized into 3 groups: less than 20 years, 20 to 34 years, and more than 34 years. Maternal educational attainment has been shown to have a profound effect on number of births and risk of adverse birth outcomes¹³; therefore, it was categorized into 3 groups: less than high school (≤9 years of formal education), high school (10–12 years), and some college or beyond (>12 years).

Because income data are not available in vital statistics records, maternal education was used as a proxy for socioeconomic status, as is common in studies that use data from vital statistics records.^15,18 Infants born to unmarried mothers have been found to have a higher LBW rate¹⁰; therefore, maternal marital status was included in the analysis. Adequacy of prenatal care was coded according to standard methods to include the month prenatal care began, number of prenatal visits, and gestation period.¹⁹

Children's characteristics, including gestational age, gender, live birth order (first, second, third, and above), were included because female infants on average are smaller than male infants and previous research²⁰ has shown that first-born children have a higher risk of LBW. Mother's nativity was included because previous studies have found that infants of US-born mothers are at higher risk of death.²¹ Maternal smoking and drinking behaviors during pregnancy have repeatedly been linked to adverse birth outcomes. Outcomes of interest included birthweight, LBW rate, 5-minute Apgar score, and 1-year infant mortality.

Propensity Score

I used propensity score analysis to adjust confounding factors across groups. Demographers have long discussed how to decompose the mean difference in the outcome variable between groups into the portion attributable to differences in the distribution of confounding variables and differences in the key explanatory variable of interest,²² and propensity scoring is one available nonparametric method to achieve this goal. A propensity score is a model-based predicted probability of an individual's membership in the reference group, given his or her covariates (i.e., characteristics). Rosenbaum and Rubin²³ demonstrated that the propensity score captures all of the variance in the covariates necessary for adjusting between-group comparisons to match 2 different groups on the basis of this single variable (the propensity score).

I estimated the propensity scores using a nonparametric regression technique that minimizes imbalance in covariate distributions between groups,²⁴ then used the estimated propensity score to weight each individual infant in the comparison group (e.g., White–White) to match the reference group (White–Black). In this approach, an individual more similar to those in the reference group receives a higher propensity score and, thus, higher weight. This method creates a hypothetical group of study participants who are similar to the members in the reference group in terms of all characteristics other than actual group membership (i.e., maternal and paternal race/ethnicity). The benefit of using propensity score weighting is that after weighting the empirical distribution of all other factors (e.g., parental biofeatures, socioeconomic characteristics) should be nearly identical across groups. Therefore, the only observed attributes that differ between the 2 groups, by design, should be race/ethnicity and birth outcomes of interest.

For example, by weighting the White–White group, I could examine the birth outcomes of infants born to 2 White parents as though they had exactly the same demographic and socioeconomic characteristics as those born to a White mother and a Black father. All covariates had similar distributions in both groups as a result of the propensity score weighting, so any remaining difference in birth outcomes between groups (i.e., White–White and White–Black), should be attributable to paternal race/ethnicity, assuming most important risk factors were captured.

In this analysis, infants born to White mothers and Black fathers were the reference group because Blacks have long been the focus of birth-outcome disparity research, and all other racial combinations were comparison groups that were individually weighted to match the reference group. For instance, infants born to a White mother and a White father were one of the comparison groups. White–White infants who had background covariates similar to those of White–Black infants received higher propensity scores and weights. Conversely, White–White infants who had characteristics very dissimilar to those of White–Black infants received lower propensity scores and weights. Other comparison groups included White–Mexican, White–American Indian, White–Puerto Rican, White–Central/South American, and White–unreported.

TABLE 2

TABLE 3

TABLE 4

Maternal Race and Infant Race

In general, I found substantial variation in birth outcomes within the group of infants with White mothers and fathers of different racial/ethnic groups. This is interesting because it shows that the common practice of using maternal race/ethnicity to refer to the infant's race/ethnicity, regardless of father's race/ethnicity, can be problematic. For example, it is not uncommon for a study to refer to infants of White mothers as “White infants,” even though “White infants” may imply that the fathers are White. In this study, I demonstrated that infants of a White mother and a White father, the real “White infants,” have the better birth outcomes than do those infants of a White mother and a non-White father. Therefore, the practice of using “White mother” to refer to White infants will yield lower estimation of the birth outcomes because there are infants of non-White fathers in the sample.

The infants in the White–White group had the most-advantaged birth outcomes, followed by infants in the 3 Hispanic-father groups. Infants in the White–Black group had the second-most-disadvantaged birth outcomes; the differences in birth outcomes between White–Black and White–White infants were statistically significant: White–White infants had a 2% (70 g) higher average birthweight, 26% lower LBW rate (4.64% vs 6.26%), and 39% lower infant mortality rate (0.43% vs 0.71%) than did White–Black infants. Infants in the White–unknown group had the most-disadvantaged outcomes in each category. These heterogeneities within White mothers show that the common practice of using maternal race/ethnicity to refer to the race/ethnicity of the infant is problematic: White–White infants had the best birth outcomes among the groups studied, so any other paternal race/ethnicity pulls down the averages for all White mothers. That is, the birth outcomes of White–White infants are actually underestimated by researchers who use mothers’ race/ethnicity to refer to infants’ race/ethnicity, and thus, the racial/ethnic disparities between White and any other race/ethnicity may be underestimated accordingly as well.

The use of maternal race to refer to infant's race in research has already been recognized as a potential source of problems. In 1997, to reflect “the increasing diversity of our Nation's population that has resulted primarily from growth in immigration and in interracial marriages,”¹⁶ the Federal Office of Management and Budget revised the standards for how the federal government would collect and present data on race/ethnicity. For example, Census 2000 included as many as 63 possible combinations of race.²⁵ Despite this effort, the use of maternal race/ethnicity to refer to a child's race/ethnicity is quite common in most research and many data-collection efforts, such as the National Center for Health Statistics and research that uses data from the National Center for Health Statistics, because “[the National Vital Statistics System] is based on data collected by the states, [and it] will not be fully compliant with the new standards until all of the States revise their birth certificates to reflect the new standards.”^26(p4) This approach, as discussed previously, may yield underestimation; therefore, before the 1997 Office of Management and Budget standards are widely implemented and accepted, researchers need to be clear and cautious when using maternal race/ethnicity to refer to child's race/ethnicity.

Paternal Racial Effect

I also observed that White women with White partners were different in many ways from those with non-White partners. On the whole, the White–White group had the most advantaged parental and children's characteristics, whereas the White–Black and White–unknown groups had the most disadvantaged parental characteristics. These patterns are consistent with previous studies of intermarriage in United States, such as Fu's 2001 study.⁷ Because we did not have socioeconomic characteristic information of fathers in the data, we could not test the status-exchange hypothesis (i.e., we don't know whether White male partners of White women were socioeconomically better off than non-White partners of White women), but this data analysis is consistent with the in-group preference hypothesis. The observed difference implies that, for providers and researchers, knowing male partners may help identify potential risks.

When I compared infants in the weighted groups to those in the White–Black group, only very small differences in outcomes remained across groups that may be attributed to paternal race/ethnicity or omitted variables such as income. These results showed that the disparities in birth outcomes among subgroups of White mothers can largely be attributed to nonracial parental characteristics. The small paternal racial effect on birth outcomes found here is a result, in large part, of the fact that mothers play a more important role than do fathers in the course of pregnancy, and therefore, mother's race and ethnicity has more influence on infant outcomes, as Table 4 shows.

As discussed previously, the White– unknown group is a category of its own. With the most-disadvantaged covariates, we unsurprisingly found that this group's outcomes were the worst. Even after we applied the propensity-score weighting to match the White–unknown group to the White–Black group, which improved all the covariates of the weighted White–unknown group to the same level of the White–Black group, the outcomes of White–unknown infants still remained significantly worse than those of other groups. Clearly, the unreported father is a proxy for more-noteworthy factors, because if unreported fathers were merely missing from certificates, their infants’ outcomes should not be so much worse.

Further research is needed to investigate the reasons why the father's information is missing. In particular, we want to distinguish fathers just missing on birth certificates for some reason from those completely missing from the child's life. For example, 7% of mothers were married in the group of unreported fathers, and those fathers are probably not entirely missing from the child's life. This result confirms findings by Gould et al., who argued that incomplete birth certificates provide an important marker for identifying high-risk women and vulnerable infants. They also pointed out that data “cleaning” in analysis might result in the removal of those mothers and infants at risk.²⁷ In this study, I found that unreported fathers were perhaps the most vital marker for the identification of infants at risk.

Although the exact pathways of how unreported fathers would affect infant health at birth are not definite, a possibility is that infants in this group are unintended, or even unwanted, births and they are likely to receive less care than intended infants, both during pregnancy and after birth. In this study, female partners of unreported fathers were disproportionately young (29% were younger than 20 years), received less-adequate prenatal care, and had a strikingly high smoking rate (39%), which indicates the mothers may not have been prepared to get pregnant. The 1995 National Survey of Family Growth found that up to 31% of all births had not been intended, including 22% mistimed and 9% unwanted births.²⁸ Unintendedness itself poses an added and independent risk factor to both women and infants and often precludes people from participating in preconception risk identification.^29,30

Implications for Public Health

The results of this study have multidimensional implications for public health. First, counter to the common conception that White mothers are at lower risk for poor infant health, I found that there were great variations within the White mother group, especially in education and behaviors. This implies that prenatal interventions need to pay more attention to the disadvantaged mothers within all racial/ethnic groups, even when the particular race/ethnicity is normally not considered at risk. As Hogue and Vasquez argued, it is time for health policymakers to prioritize integration of the concerns of all pregnant women, not just those of minorities.³¹

Secondly, the father deserves more attention from researchers. Not only can the father's characteristics help to identify potential risks, but his behaviors also may affect the mother's behaviors. For example, some previous research has shown that, regardless of marital status, a woman's substance use during pregnancy is highly correlated with both her partner's substance use and the degree of emotional supports he provided.³² Unfortunately, no information on male partners’ behaviors in the data set means that such an analysis is beyond the scope of this study.

Finally, further research is needed to investigate why so many fathers are unreported and how unreported fathers may affect birth outcomes. Identifying the reasons that fathers are missing on the birth certificate is the first step. In parallel with further research efforts, policies and programs need to be carefully designed to address how to improve outcomes of unintended pregnancy.

This study was supported by the Pardee Rand Graduate School Dissertation Award.

The author gratefully acknowledges the generous comments and suggestions from Robert A. Hummer and Rand researchers Brian K. Finch, Greg Ridgeway, Julie DaVanzo, and Becky Kilburn.

No protocol approval was needed for this study, because data were deidentified.