Low birth weight and preterm births vary by state, and Black mothers typically face twice the risk that their white counterparts do. This gap reflects an accumulation of psychosocial and material exposures that include interpersonal racism, differential experience with area-level deprivation such as residential segregation, and other harmful exposures that the authors refer to as “institutional” or “structural” racism. The authors use logistic regression models and a dataset that includes all births from 1994 to 2017 as well as five state policies from this period—Aid to Families with Dependent Children/Temporary Aid for Needy Families, housing assistance, Medicaid, minimum wage, and the earned income tax credit (EITC)—to examine whether these state social policies, designed to provide a financial safety net, are associated with risk reduction of low birth weight and preterm birth to Black and white mothers, and whether variations in state generosity attenuate the racial inequalities in birth outcomes. The authors also examine whether the relationship between state policies and racial inequalities in birth outcomes is moderated by the education level of the mother. We find that the EITC reduces the risk of low birth weight and preterm birth for Black mothers. The impact is much less consistent for white mothers. For both Black and white mothers, the benefits to birth outcomes are larger for mothers with less education.
In the United States, Black-white gaps in birth weight and preterm birth are a persistent and tragic feature of a political and economic system that produces and reproduces health inequities—differences in health outcomes that are avoidable and preventable and are therefore unjust (Whitehead 1992). Black mothers have a nearly twofold greater risk of low birth weight (infants born <2500 g) and preterm birth (infants born before 37 weeks of gestation) compared to whites (Collins et al. 2000; Rosenthal and Lobel 2011). Even with comparable levels of formal education, Black women face greater risk, and the most highly educated Black mothers have higher levels of preterm and low-birth-weight infants than white mothers with the least amount of education.
While the causes of preterm birth and low birth weight are not fully understood, they are thought to reflect a complex interaction of the mother's age; genetic factors; behavioral factors, especially smoking and substance abuse; and (lack of) prenatal care. Figure 1 shows that for both outcomes, there is a graded pattern for Black and white women according to levels of educational attainment. This is consistent in analyses that examine socioeconomic statistics at the individual and area levels (Blumenshine et al. 2010). However, as noted before, at comparable points along the socioeconomic ladder, Black infants are at a higher risk (McGrady et al. 1992). The Black-white gap in infant outcomes does not reflect mysteries of DNA, but rather the cumulative effects of racism.1 “Black” identity is not a discrete genetic category that captures a unique profile of medical risk but rather a lived, contingent, and contextual experience, which, at comparable points along the socioeconomic ladder, typically means having less access to the types of resources that promote health and more harmful psychosocial and material exposure.
Both racial (ethnic) and class inequalities thus matter to health outcomes, as these categories are “mutually constitutive” (Kawachi, Daniels, and Robinson 2005). Essentialist notions of a Black identity (or “African” or “Negro” in earlier periods) originated and evolved as ideological justification of US slavery and, in the post-Reconstruction and post–Jim Crow eras, as rationale for systems of low-wage, exploitative labor regimes (Fields 1990; Smedley 2012).
Studies of Black-white inequalities in birth outcomes in the United States have increasingly pursued analyses at two levels, starting with an assessment of risk factors for mothers (age, education, smoking, access to prenatal care, etc.) while also considering the context of racism (interpersonal and structural) in which births happen. David and Collins (1997, 2007) advanced this perspective in an important set of articles that challenged the genetic interpretation of Black-white disparities in birth outcomes by showing evidence that perceived discrimination predicted low birth weight. Subsequent studies have analyzed exposure to interpersonal discrimination in different domains, and many find an association with perceived racism that, over time, affects birth outcomes.2
By looking at individual-level risk factors in the context of neighborhoods or other geopolitical units of analysis, researchers have found that different metrics of concentrated disadvantage, or “deprivation,” matter to birth outcomes and differ for Black and white women. Residential segregation, local (or national) labor markets, poverty rates, environmental exposures such as lead and air pollution, and racial attitudes are all associated with racial disparities in health, beyond what individual risk factors alone would predict (Burris and Hacker 2017; Chae et al. 2018; Orchard and Price 2017). All of these factors capture elements of “institutional” or structural racism, and a number of articles have attempted to operationalize the concept and measure the effects on birth outcomes.3
In addition to the foci on racism at the individual and “structural” levels, a largely separate research stream has explored potential social policy and income-support mechanisms on the assumption that these might disrupt the pathway between lower socioeconomic status (SES) and health (Hoynes, Miller, and Simon 2015; Riley et al. 2021; Strully 2011). Studies of antipoverty polices such as Aid to Families with Dependent Children/Temporary Aid for Needy Families (AFDC/TANF) and Medicaid participation in the Special Supplemental Nutrition Program for Women, Infants, and Children (Kowaleski-Jones and Duncan 2002) have produced results that we might expect: social policies improve birth outcomes (Almond, Hoynes, and Schanzenbach 2011). A number of articles have found that the earned income tax credit (EITC) is associated with better health among infants (Hamad and Rehkopf 2015; Hoynes, Miller, and Simon 2015; Markowitz et al. 2017; Strully, Rehkoph, and Zuan 2011; Wagenaar et al. 2019) and that higher minimum wages are associated with better birth outcomes.
Our contribution builds on this extant multidisciplinary literature and on an insight drawn from lessons of African American political history. From the New Deal era to the present, a broad array of African American thinkers and activists who pushed for racial equality emphasized a multifaceted agenda: broad and equitable social welfare provision for all, full employment, and antidiscrimination law and public policy. Mainstream civil rights organizations, labor unions, and more radical groups insisted that federal provision of social welfare and employment were preconditions to advancing racial equality. Indeed, the 1963 March on Washington was for “Jobs and Freedom” and called for full employment, a New Deal style of public works, higher minimum wages, and school desegregation, in addition to support for the Civil Rights Act (Hamilton and Hamilton 1997; Reed 2020a; Reed 2020b).
They did so with good reason. Federal power and authority matter more in terms of the social determinants of health in the United States and in addressing racial health disparities because macroeconomic policies, health care access, social policies of various kinds, workers' rights, the making and enforcement of antidiscrimination law, and tax policies that promote income (and wealth) redistribution are all most effectively accomplished as implementations of national policies. Indeed, when the federal government addressed racial and class inequalities in the 1960s through the War on Poverty and Great Society programs, and coupled those efforts with antidiscrimination law, health disparities narrowed, especially for Black women and infants (Almond, Chay, and Greenstone 2007; Hahn et al. 2018; Kaplan, Ranjit, and Burgard 2008; Krieger et al. 2013).
By 1980 and the election of Ronald Reagan as US president, the nation had moved away from 1960s-era social and economic policies and in the direction of neoliberal ones. This meant relatively flat minimum wages, less federal protection for civil rights and workers' rights, deregulation of markets, and trade deals that facilitated corporate mobility to lower-wage economies, all of which resulted in widening economic inequality and persistent racial disadvantage (Osterman 1999; Kalleberg 2011).4
During this time, states experimented with and sometimes extended a range of social welfare policies that impact health outcomes (Howard 2007)5, but policy design and generosity have varied significantly across the 50 states (Howard 2007; Campbell 2014). Bruch, Meyers, and Gornick (2018) looked at 10 federal-state programs and found consistent cross-state inequalities in provision. They also found that, after the Personal Responsibility and Work Opportunity Reconciliation Act of 1996, the federal government's devolution of authority over the administration of these programs increased inequalities in cross-state provision. This meant less direct cash assistance to the poor and the establishment of work requirements and punitive sanctions for not complying (Soss, Fording, and Schram 2011).6
We set out to examine the question of whether more generous social welfare provision among the 50 states might have an impact on Black and white low birth weight and preterm birth as well as the Black-white gap, particularly in the period following the Personal Responsibility and Work Opportunity Reconciliation Act of 1996. Given that birth outcomes follow a socioeconomic gradient, we also wondered whether the impact of social welfare policies differs by level of formal education for Black and white mothers.
Three articles have covered similar ground. Hoynes, Miller, and Simon (2015) used a difference-in-difference model to explore how the state expansion of the state EITC in 1993 impacted rates of low birth weight for white, Black, and Hispanic mothers with no more than a high school education. The expansion lowered rates of low birth weight for all mothers; however, the effects were strongest for Black mothers and weakest for white mothers. Wagenaar and colleagues (2019) categorized the EITC into four types based on the amount of the payment (low vs. high) and whether payments are refundable. They found that the strongest form of EITC (high levels with a refund) were associated with decreased rates of low birth weight and longer gestation periods for all mothers; but the effects were stronger for Black mothers than for white mothers. As with Hoynes, Miller, and Simon, their sample was women with no more than a high school diploma. Wehby, Dave, and Kaestner (2019) examined how the state minimum wage impacts birth weight and gestation for white and nonwhite mothers. They found that a minimum wage increase was associated with decreased rates of low birth weight for both white and nonwhite mothers, although the effects were stronger for nonwhite mothers. They also found that minimum wage increases were associated with lower rates of preterm birth for white mothers but not nonwhite mothers.
Our research differs from the previous studies in important ways. The first is conceptual. We understand the persistent racial gap in birth outcomes as the result of racism—interpersonal and structural—and we consider social welfare provision as a potential intervention. Second, we consider a broader set of policies that include safety net programs such as AFDC/TANF along with two that increase income through labor force participation—the minimum wage and the EITC, which we examine simultaneously.7 Fourth, we operationalize the EITC in actual dollar amounts available to residents of specific states. Finally, and to weigh the impact of social policies within race, we stratify both the mother's race and her level of education.
Our study takes advantage of a dataset that allows us to include outcomes for every US live birth between 1994 and 2017—more than 60 million in total—along with detailed information about the mothers, including race/ethnicity, age, education level, prior pregnancies, prenatal care, and smoking behavior. We have data from the same period on state policies that aim to simultaneously mitigate economic inequality and assist low-income families with meeting basic economic needs. These include the EITC, the minimum wage, AFDC/TANF coverage, housing assistance, and Medicaid/Children's Health Insurance Program (CHIP). This interval of time captures significant changes in the US political economy in terms of tax policy, trade, workers' rights, minimum wages, and social welfare provision, and more specifically in the policies mentioned above. For instance, AFDC, one key element of the economic safety net, underwent dramatic revisions in 1996 with the introduction of the Personal Responsibility and Work Opportunities Reconciliation Act. The EITC expanded during this period.
Data and Sample
We use the National Vital Statistics System natality files from 1994 to 2017, compiled by the National Center for Health Statistics at the Centers for Disease Control and Prevention. These data, which are drawn from United States Standard Certificates of Live Birth, provide individual-level information on all infants born in the United States during this period. Demographic characteristics of the mother and infant as well as information on prenatal care and smoking during pregnancy are included in the dataset. We restrict the sample to singleton births to non-Hispanic white or non-Hispanic Black mothers. Mother's race is missing for less than 0.1% of the sample.8 We include data from the 50 states but exclude Washington, DC.
We originally planned to use a sample consisting of all live births from 1980 to 2017, with state policy variables measured in each year from 1980 to 2017. However, because of the high correlation between the EITC and calendar year prior to 1994, we truncated our sample to produce reliable estimates for the EITC. It should be noted that when using the sample from 1980 to 2017, the findings for the other state policy variables were very similar to the results presented in this article.9
Our state-level policy information is drawn from several sources. We use information on EITC and minimum wage policies for 1994–2017 from the University of Kentucky Poverty Research Center's National Welfare Data set. We use data from the Annual Social and Economic Supplements to the Current Population Survey for 1992–2017 to create our additional state policy measures. Finally, we use data compiled by the US Department of Commerce and the US Census Bureau on state revenue and expenditures from 1994 to 2017.
We use two outcome variables: the first is a binary indicator of whether the infant was low birth weight (<2500 g) or not low birth weight (≥2500 g), and the second is a measure of whether the birth was preterm. We define preterm birth as gestation of fewer than 37 weeks.
We measure the education of the mother as a series of five categories: did not complete elementary school, completed elementary school but did not graduate from high school, high school diploma, some college education, and bachelor's degree or more. We also include measures for the marital status of the mother (married vs. not married), the mother's age in years, whether this is the first live birth for that mother, and the sex of the infant. To allow for a nonlinear relationship between mother's age and our outcomes, we include both a linear and a squared term for the mother's age. In addition, we include a measure for the extent of prenatal care the mother received. Prenatal care is coded as adequate if this care began during the first two trimesters and inadequate otherwise. We also include a variable for whether the mother smoked during pregnancy. Finally, we include variables for the month and calendar year of the birth. To allow for flexibility in changes over time, month and year are modeled as nominal categorical variables (e.g., month and year fixed effects).
State-Level Policy Variables
We include five state-level policy variables. All of these are time-varying (annual) measures for each year from 1994 to 2017 and are based on the state in which the mother resided at the time of the birth. Each of these policies is a key element of the social safety net designed to help low-income families meet basic economic needs.
▪ Minimum wage. This measure is the state minimum wage adjusted for annual inflation by the region-specific Consumer Price Index (CPI).
▪ EITC. This measure is the maximum level of earned income tax credit, including both the federal and the state-specific portions, that an individual with three dependents is eligible for in each state in each year. Our EITC measure is adjusted for annual inflation by the region-specific CPI provided by the Bureau of Labor Statistics. The base CPI year is 1983. The EITC is measured in $100s.
▪ AFDC/TANF coverage. This measure is a ratio. The numerator is the number of households in the state receiving AFDC/TANF, and the denominator is the number of households with children younger than the age of 18 in the state whose income falls below the federal poverty level.
▪ Housing assistance. This measure is a ratio. The numerator is the number of households in the state receiving a rent subsidy from the federal, state, or local government. The denominator is the number of households in the state with incomes below the federal poverty level.
▪ Medicaid/CHIP. This measure is a ratio. The numerator is the number of individuals in the state receiving Medicaid or CHIP. The denominator is the number of individuals in the state living in households with incomes below the federal poverty level.10
Additional State-Level Variables
To capture the general macroeconomic context, we include a measure of the state poverty rate and state unemployment rate in each year. We also include two variables to capture state levels of spending that benefit all individuals but are not specifically part of the social safety net: annual measures of state health/hospital expenditures and education expenditures. The education measure includes both K–12 and higher education spending. Health/hospital and education expenditures are measured in $1,000s per capita. We also include annual measures of the fraction of the population aged 18 years or younger and the fraction of the state population aged 65 or older. We include these variables as they may influence the level of spending on health and education. Finally, we include a time-invariant indicator (0/1) variable for each state (e.g., state-level fixed effects) that allows us to capture any time-invariant, unobserved, state-level characteristics that may influence gestation and birth weight.
For each model, we exclude cases with missing data on the outcome variable (low birth weight or preterm birth). We also restrict the analysis to cases with complete information on mother's race, age, education level, marital status, child's sex, and live birth order. Missing data on most of these characteristics is minimal. There is no missing data on mother's age or the sex of the infant. Marriage, live birth order, gestation, low birth weight, and preterm birth are missing for less than 1% of observations for both non-Hispanic Black and non-Hispanic white mothers. Education is missing for a slightly larger percentage of the sample: 3.1% of non-Hispanic Black mothers and 2.3% of non-Hispanic white mothers. Prenatal care received is missing for 4.5% of non-Hispanic Black mothers and 2.3% of non-Hispanic white mothers. We include cases with missing data on prenatal care along with an indicator variable for missing prenatal care information. Data on smoking during pregnancy is missing for 3.9% of non-Hispanic Black mothers and 5.6% of non-Hispanic white mothers. We include cases with missing data on smoking and also include an indicator of missing data on smoking in our models. In terms of the policy data, New Hampshire in 2011 and Rhode Island in 2015 are missing the AFDC/TANF coverage measure because of incomplete information on the number of households with children whose income is below the federal poverty level. Missouri in 1999 is missing data on housing assistance coverage because of incomplete information on rent subsidies. We exclude observations from the aforementioned states in the years in which they are missing policy data.
We estimate logistic regression models for preterm birth and low birth weights separately for non-Hispanic Black and non-Hispanic white mothers. These models allow the impact of both state policy variables as well as individual covariates to vary by race. We then further stratify the models by education level, separately estimating models for mothers of each race who do not have a high school diploma, who have a high school diploma (no college education), who have some college education, and who have a bachelor's or graduate degree. To interpret the magnitude of the results, we use the coefficients from these models to microsimulate a series of predicted probabilities for each outcome, where we allow values for the five policy variables of interest to vary.11 We estimate predicted probabilities at the 1st percentile, 50th percentile, and 99th percentile of the values on each policy variable. Percentiles are derived from our combined sample of non-Hispanic Black and non-Hispanic white mothers from 1994 to 2017. We use STATA version 16 for all analysis.
Trends over Time in Low Birth Weight and Preterm Births by Race
Figures 2 and 3 show trends over time in rates of low birth weight and preterm birth respectively. There were only minor fluctuations in rates of low birth weight from 1994 to 2017, with no substantial trend over time. Low birth weight rates for non-Hispanic Black mothers were consistently more than twice the rates for non-Hispanic white mothers. The patterns for preterm birth are somewhat different. Between 1994 and 2017 the rates of preterm birth for non-Hispanic Black mothers dropped by about 2 percentage points (from 17.0% to 14.9%). The rates did rise slightly (from 14.4% to 14.9%) between 2015 and 2017; there is no way knowing if this trend will continue. For non-Hispanic white mothers, the rates of preterm birth rose and then fell during this time period; rates in 2017 were half a percentage point higher than in 1994 (8.5% vs. 8.0%). The gap between non-Hispanic Black mothers and non-Hispanic white mothers narrowed during this time period. Nevertheless, in 2017, rates for non-Hispanic Black mothers were still 1.75 times as high as those for non-Hispanic white mothers.
We see from table 1 that the rates of low birth weight and preterm birth are higher for non-Hispanic Black mothers than for non-Hispanic white mothers. Educational attainment and marriage rates are higher for non-Hispanic white mothers. Relative to non-Hispanic Black mothers, non-Hispanic white mothers are more than twice as likely to be married, are 11 percentage points more likely to have at least a high school diploma, and are more than twice as likely to have at least a bachelor's degree. The great majority of both non-Hispanic Black and non-Hispanic white mothers receive adequate prenatal care, although rates are twice as high for non-Hispanic white mothers. Finally, non-Hispanic Black mothers are about 3 years younger on average and are just over 5 percentage points less likely to smoke.
Bivariate Regression Models
We first estimate bivariate logistic regression models for low birth weight and preterm birth, each model containing one of our five key policy variables. Models are estimated separately for non-Hispanic Black and non-Hispanic white mothers. These results are presented in appendix table A1. For both outcomes and both samples of mothers, increases in each policy variable are associated with declines in both preterm birth and low birth weight. All results are statistically significant at p < .001, which is what we would expect. These models are a useful starting point, but they include no individual covariates or year or state fixed effects. Therefore, we cannot rule out the possibility that associations between policies and outcomes are the result of other state characteristics or characteristics of the individuals living within these states that are correlated with both the policies in question and the birth outcomes.
Multivariate Regression Models
We then estimate models including both non-Hispanic white and non-Hispanic Black mothers to show the overall racial disparities in low birth weight and preterm birth, net of any differences in demographic characteristics, prenatal care, and social welfare policies in the states in which these mothers live. As table 2 shows, the likelihoods of low birth weight and preterm birth are statistically significantly higher for non-Hispanic Black mothers as compared to non-Hispanic white mothers. The model-predicted probabilities of low birth weight are 10.3% for non-Hispanic Black mothers and 5.2% for non-Hispanic white mothers. The model-predicted probabilities of preterm birth are 14.0% for non-Hispanic Black mothers and 9.0% for non-Hispanic white mothers. These regression models do not allow for the fact that the impact of the policy variables and individual covariates may vary by the mother's race. To incorporate this difference, we then estimate models separately for non-Hispanic Black and non-Hispanic white mothers.12
Table 3 shows the results for models of low birth weight and preterm birth, estimated separately for non-Hispanic Black and non-Hispanic white mothers. For both groups, the impact of individual covariates is in the expected directions. For both groups of mothers, low birth weight and preterm birth are more frequent for mothers at older ages. The risk of low-birth-weight births rises with age from age 11 for non-Hispanic Black mothers and from age 8 for non-Hispanic white mothers, effectively the earliest years for which pregnancy can occur. Preterm births rise with age from age 22 for non-Hispanic Black mothers and age 20 for non-Hispanic white mothers.13 Risk of low birth weight and preterm birth are also higher for unmarried mothers, mothers who have less formal education, mothers who have received inadequate prenatal care, and mothers who smoke during pregnancy. Low birth weights are more frequent, although preterm births are less frequent for female infants.
The impacts of the state-level controls are less consistent. For instance, educational expenditures are associated with a decrease in low birth weight and preterm birth for non-Hispanic Black mothers, but not for non-Hispanic white mothers. Health expenditures are associated with lower rates of preterm birth for both samples of mothers and with low birth weight for non-Hispanic white mothers. Health expenditures are, however, associated with an increase in low birth weight for non-Hispanic Black mothers. It is important to note that this measure not only captures state investment in health care but also represents the number of sick people in the state who may need care. This may help to explain the positive association with low birth weight.
The results for our key state policy variables of interest are as follows. Increases in the level of the EITC and AFDC/TANF coverage are associated with a statistically significant lower risk of low birth weight and preterm birth for non-Hispanic Black mothers. For non-Hispanic white mothers, the EITC is associated with a statistically significant lower risk of preterm birth only. AFDC/TANF coverage is actually associated with an increase in low birth weight rates for non-Hispanic white mothers; there is no association for preterm birth. The other three policy variables also have scattered statistically significant associations with birth outcomes that are in the opposite direction from what we would expect: Medicaid/CHIP coverage is associated with a higher rate of low birth weight and preterm birth for non-Hispanic white mothers. Housing assistance coverage is associated with a higher rate of low birth weight and preterm birth for non-Hispanic Black mothers. The minimum wage is associated with higher rate of preterm birth for non-Hispanic Black mothers.
Because of a sample size of more than 18 million non-Hispanic Black mothers and more than 70 million non-Hispanic white mothers, even trivial associations will be statistically significant. All results, especially those in unexpected directions, should be understood in this context.14 We thus use predicted probabilities to examine the magnitude of these associations.
Table 4 shows the values for each state policy variable at the 1st, 50th, and 99th percentile. These percentiles are drawn from the entire sample of non-Hispanic Black and non-Hispanic white mothers from 1994 to 2017. We also calculated the percentiles using mothers of all race and ethnicities; the results were effectively identical. Table 5 shows the predicted probabilities of low birth weight and preterm births for non-Hispanic Black and non-Hispanic white mothers at the 1st, 50th, and 99th percentiles for each policy variable. All state policy variables are included for the sake of completeness; the statistical significance of the association between the state policy variable and the birth outcome is shown in the right column of the table. As table 5 shows, other than for the EITC, the impact of the state policies on birth outcomes are generally very small.
For non-Hispanic Black mothers, the impact of the EITC on both low birth weight and preterm birth is substantial. As table 5 shows, at the 1st percentile of the EITC, the model-predicted low birth weight is 11.8%; this falls to 10.9% for the 99th percentile of the EITC. Preterm births are 16.2% at the 1st percentile of the EITC, dropping to 14.9% by the 99th percentile. As table 6 shows, these changes are approximately comparable to the impact of getting a high school diploma, obtaining some college education, and giving birth five years earlier in life.
For non-Hispanic Black mothers, none of the other state policy variables are associated with a change of more than 3 percentage points in predicted rates of low birth rate or preterm birth. Several of these associations are not statistically significant, as noted by (NS) on the table. For non-Hispanic white mothers, no state policy variable is associated with a change of more than 1 percentage point in low birth weight rates or a change of more than 2 percentage points in preterm birth rates. Most of these associations are not statistically significant.
Regression Models: Stratified by Education
To further disentangle the impact of public policies on racial and socioeconomic disparities in birth outcomes, we estimated models separately for non-Hispanic Black and non-Hispanic white mothers with less than a high school diploma, a high school diploma, some college, and a bachelor's degree or more. Model results are shown in appendix tables A3–A6. Results for the EITC are illustrated in Figure 4 (low birth weight) and figure 5 (preterm births).
Table A3 shows that for non-Hispanic Black mothers at all levels of education below the bachelor's degree, as the EITC increases there is a statistically significant decline in the likelihood of low birth weight. For non-Hispanic Black mothers with a bachelor's degree, there is no statistically significant association between the EITC and low birth weight. We see from figure 4 that the magnitude of this association is strongest among mothers with lower educational attainment. For non-Hispanic Black mothers without a high school diploma, model-predicted rates of low birth weight are 13.8% at the 1st percentile of EITC benefits; this falls to 12.6% when the mother lives in a state with EITC benefits at the 99th percentile. While non-Hispanic Black mothers with a high school diploma have lower rates of low birth weight overall, the magnitude of the association with the EITC is similar to that for mothers with no high school diploma. At the 1st percentile of the EITC, the predicted rate of low birth weight is 12.2%; this falls to 11.1% for the 99th percentile of the EITC. For non-Hispanic Black mothers with some college education, the corresponding drop is only 0.7 percentage points (from 10.9% to 10.2%). For non-Hispanic Black mothers with a bachelor's degree, the predicted rate of low birth rate is 8.9%, regardless of the value of the EITC.
The results are very different for non-Hispanic white mothers. As table A4 shows, for those who have at least a high school diploma, there is not a statistically significant relationship between the EITC and low birth rate. For non-Hispanic white mothers with no high school diploma, a more generous EITC benefit is associated with an increase in the rate of low birth weight and this relationship is statistically significant. As figure 4 shows, at the 1st percentile of the EITC, the predicted rate of low birth weight is 8.0%; this increases to 8.6% at the 99th percentile of the EITC.
Figure 4 also illustrates that non-Hispanic Black mothers with a bachelor's degree have higher model-predicted rates of low birth weight than non-Hispanic white mothers with no high school diploma. Even at the 99th percentile of the EITC, non-Hispanic Black mothers at any education level have higher rates of low birth weight than all non-Hispanic white mothers.
Results for preterm birth both mirror and diverge from those for low birth weight. Table A5 shows that for non-Hispanic Black mothers with all educational attainment levels below the bachelor's degree, as the EITC increases there is a statistically significant decline in the likelihood of preterm birth. Again, the exception to this is non-Hispanic Black mothers with a bachelor's degree, for whom there is not a statistically significant association between the EITC and preterm birth. However, the magnitude of the association between EITC and preterm birth is similar for all non-Hispanic Black mothers with less than a bachelor's degree. As figure 5 shows, for non-Hispanic Black mothers with no high school diploma, the predicted rate of preterm birth is 18.9% when at the 1st percentile of EITC benefits. This falls to 17.3% at the 99th percentile of EITC benefits. The trend for mothers with a high school diploma is very similar: 16.8% at the 1st percentile of the EITC and 15.1% at the 99th percentile of the EITC. For non-Hispanic Black mothers with some college education, the corresponding drop is slightly smaller: from 15.0% at the 1st percentile of the EITC to 13.7% at the 99th percentile of the EITC.
Table A5 shows that for non-Hispanic white mothers, the only statistically significant association between EITC and preterm birth is for mothers with a high school diploma (see table A6). The association between the EITC and preterm birth is modest for these mothers; at the 1st percentile of the EITC, the predicted rate of preterm birth is 10.0%; this falls to 9.4% at the 99th percentile of the EITC.
As with low birth weight, non-Hispanic Black mothers with a bachelor's degree have higher model-predicted rates of preterm birth than non-Hispanic white mothers with no high school diploma, illustrating that racial disparities are a stronger predictor than socioeconomic disparities in terms of influencing birth outcomes.
Discussion and Conclusion
We started with the puzzle of Black-white disparities in preterm birth and low birth weight and the understanding that SES did not adequately capture the gap because of interpersonal and structural (or “institutional”) racism. We drew from civil rights history and captured a long-standing view that tackling racial injustice required federal and universal social welfare provision, full employment, and robust antidiscrimination law and policy. We then considered more than 60 million births over the course of three decades in relation to a specific set of federal policies that states administer and (sometimes) enhance.
Although our research uses a very robust data set, spanning 23 years of policy changes and including every US birth during that period, there are some limitations to this data. First, the individual-level data tell us whether mothers had access to prenatal health care but not whether they directly benefited financially from the particular policies we modeled. Thus, our results cannot say conclusively that making use of the financial resource of a particular policy benefits the birth outcomes for individual mothers. Rather, our research suggests that having access to the financial benefits of these policies is associated with improved birth outcomes. Relatedly, we do not have information on the household income level of the mothers; thus, we rely on education as a proxy for financial resources to stratify our model by socioeconomic class. While education is generally a good proxy for socioeconomic class, it does not completely capture diversity in financial resources.
Our results showed that the EITC reduces the risk of preterm birth and low birth weight for Black infants, and this association is strongest for Black mothers with the lowest levels of education. We found no association between the EITC and low birth weight for white infants, and the association between the EITC and preterm births for white infants was very small. Other safety net policies had minimal impact on birth outcomes for both Black and white infants.
It is not clear why the EITC has a stronger impact on birth outcomes relative to other social safety net policies, but it may be because the EITC has further reach than several other policies designed to assist low-income individuals and families. For instance, while only a limited number of housing assistance vouchers are available, there is no limit on the available coverage for the EITC; all employed individuals who earn below the specified threshold receive the credit. In addition, income eligibility restrictions for AFDC/TANF are much stricter than those for the EITC. For comparison, in 2017 22,220,000 individuals received the EITC (US Internal Revenue Service 2017), and monthly caseload averages for AFDC/TANF were 1,395,637 (US Department of Health and Human Services 2017). In 2016, 4.09 million households received governmental housing assistance from the Department of Housing and Urban Development (Kingsley 2017). In 2017, approximately 75 million people received health insurance through Medicaid (Rudowitz, Garfield and Hinton 2019). These numbers are not strictly comparable, as they operate on different scales. The EITC, AFDC/TANF, and housing assistance measures are household-level measures, as in general the eligibility rules are structured so one adult in each household can receive each of the aforementioned benefits. Medicaid is an individual-level benefit and applies to children as well as adults. We did supplemental calculations using Current Population Survey data from 2017 and found that 34,187,610 individuals older than the age of 18 received Medicaid. This is still by far the policy that benefits the greatest number of people. The EITC, however, has significantly greater reach than either AFDC/TANF or housing assistance.15
Our results confirm and buttress some prior findings as well as the historical, multipronged approach to addressing racial disadvantage that civil rights, labor leaders, and other (sometimes radical) activists pushed over the course of the 20th century. We found that supplementing the income of Black mothers through EITC at least modestly addresses racial and socioeconomic disparities in birth outcomes, especially for Black women who face the most disadvantage.
These results might be important for other reasons. They hold over a period that corresponds to the end of the “liberal consensus” and corresponding changes in social and economic policy that began with the Reagan Revolution. In a period of less market regulation, less progressive taxation, erosion of workers' rights, flatter wages, and a dramatic devolution of safety net policy after 1996, our findings are consistent: more generous EITC is modestly associated with adverse birth outcomes for Black infants.
The results matter because states vary in their levels and approaches to social welfare provision, and scholars have explored potential reasons for and consequences of those differences (Franko and Witko 2017; Hertel-Fernandez 2019; Yu, Jennings, and Butler 2019). We know that states with larger Black populations make less “welfare effort” and that higher Black caseloads mean more exclusionary, punitive welfare policies and less cash assistance (Matsubayashi and Rocha 2011; Parolin 2019; Soss, Fording, and Schram 2011). We share Mettler's concern about how “submerged”16 social welfare policy (e.g., tax credits instead of direct subsidy) erodes the public's support for and understanding of governmental intervention, but in the case of the EITC that might be a virtue—it is less likely to be “racialized” than other safety net policies.
At first glance, our results might be surprising because the EITC is targeted, but not to the population that faces the greatest inequality in birth outcomes. Long-standing and persistent racial gaps in birth outcomes could suggest that only specifically targeted interventions would change the outcome. But again, the theory and historical perspective that informed our analysis would predict some protective effect, albeit a smaller effect than would result from a more robust and generous social welfare provision.
Our results were surprising with respect to white infant outcomes. The EITC does not have the same consistent associations. Why would that be? Following Williams, we speculate that the variables we use to assess SES among Black and white mothers are not necessarily commensurate (Williams 2012). A variable like “formal education” does not capture everything relevant about the lived experience, past or present, of Black and white mothers of similar SES. It does not capture wealth differences or residential patterns that might result in more toxic exposures—all of which fall under the category of structural racism. Furthermore, as a result of the cumulative and co-occurring effects of racial, class, and gendered inequality, Black mothers experience greater cumulative “wear and tear” than their white counterparts, which Arline Gernonimus calls “weathering” (Geronimus et al. 2006). That Black women start from an initially more significant economic disadvantage tied to intergenerational processes might explain why the EITC has a consistently positive impact on Black infants, but less so for whites. Part of the stronger association between the EITC and birth outcomes for Black mothers is likely also because Black mothers are starting from such an extreme disadvantage relative to white mothers vis-à-vis birth outcomes. The EITC helps to mitigate this disparity.
The corollary is that social welfare policies and spending would likely need to be significantly more generous and robust to more dramatically affect both Black and white infant outcomes, and to attenuate racial differences in birth outcomes. As of this writing, the American Rescue Plan Act is a historic policy intervention regarding safety net and income support to families and might therefore serve as a natural experiment for future investigation. Our findings, and prior research, suggest that these developments will be beneficial for infant outcomes in general and are likely to attenuate the Black-white gap in health outcomes.
Additional research might also explore why none of the other policy variables showed a consistent significant association with birth outcomes and also why the EITC had limited impact on birth outcomes for white mothers. Because of the complexity of these issues, qualitative research could include interviews of Black and white mothers of various income levels about their pregnancy and prior lived experiences that might be relevant to birth outcomes. Mothers could be selected from states with generous social policies and states with more restrictive social policies. This research could explore the extent to which mothers made use of the public policies analyzed in this article, the trade-offs they make that may impact infant health (e.g., their nutritional status while they are pregnant, the number of prenatal care visits they receive, etc.) and the ways responses differ between Black and white mothers.
Despite the remaining research questions, the findings are still clear with respect to potential interventions. Higher EITC provision is consistently good for Black mothers and their infants, particularly those who are most disadvantaged.
We thank three anonymous reviewers for their feedback. The authors contributed equally to the manuscript and are listed in alphabetical order.
For the purpose of the article, we understand “racism” to mean exposure to interpersonal, discriminatory treatment (Alhusen et al. 2016), as well as exposure to patterns of disadvantage that result in disproportionately higher levels of poverty, greater levels of unemployment, and residential segregation. In their influential book Black Power, Carmichael and Hamilton (1967) called this “institutional racism.” We take this to be synonymous with “structural racism.”
See Mutambudzi, Meyer, Reisine, and Warren (2017).
On states, unionization, and poverty, see Brady, Bakers, and Finnigan (2013).
Parolin (2019) looked at data from 2012 to 2014 and found that differences in cash assistance from TANF, attributable to Black composition in a particular state, accounted for 15% of the Black-white child poverty gap.
This is a period when the federal government's approach to providing housing to low-income Americans devolved in a way that gave states greater discretion over how public dollars were spent (Schwartz 2015).
Wehby, Dave, and Kaestner (2019) control for EITC but do not discuss the effects.
The variable for Hispanic/Latino ethnicity was missing for 0.75% of Black mothers and 1.22% of white mothers. For the analysis presented in the article, we exclude mothers for whom Hispanic/Latino ethnicity is missing. However, birth certificates where information on Hispanic ethnicity for the mother are missing may be much more likely to indicate non-Hispanic mothers. Therefore, we conducted a sensitivity analysis by estimating models where we include observations for Black and white mothers with missing data on Hispanic ethnicity. The results are extremely similar to those reported and are available on request.
Results are available on request.
We considered using a sixth variable to measure the level of AFDC/TANF benefits provided by the state. Unfortunately, this variable produced high levels of multicollinearity, rendering the results unreliable. We were forced to drop it from the model.
For these predicted probabilities, each observation retains its values on the other covariates. A predicated probability is calculated for each observation, and we then take the average of these individual predicted probabilities.
Another option is to include interaction terms between the mother's race and all other variables in the model. However, such models introduced very high levels of multicollinearity and thus were difficult to interpret.
These calculations use the formula -b/2a for the age of lowest risk, where b is the coefficient for the linear term and a is the coefficient for the squared term. This is the formula for the age of lowest risk when the squared term is positive and the linear term is negative.
The bivariate correlations among all state-level variables are presented in appendix table A1. None of the correlations exceeds .51, with the exception of a fraction of the population aged 18 and under with a fraction of the population aged 65 and older. Those two variables can be considered analogous to two parameters for a categorical variable (e.g., age). They are correlated by definition. We also conducted tests of multicollinearity for the state policy variables. None exceeded a variance inflation factor of 10.
Since the AFDC/TANF data are monthly caseload averages, it is possible these numbers underestimate the total number of families who receive AFDC/TANF at some point in the year. However, we conducted supplemental analysis using the Current Population Survey, which asks individuals to report any income from AFDC/TANF in the prior year, and the numbers were very similar to those reported by the Department of Health and Human Services.
“Submerged” policies are “those that are located in the tax code or channeled through nongovernmental organizations, making government's role in providing them less obvious” (Mettler 2020).