Why Do Black Women Have a Higher Obesity Prevalence than White Women in the United States?

Abstract

Although there are large Black-White obesity inequities among women in the United States (U.S.), the factors that explain this racialized health inequity are not well understood, most likely due to the fact that previous research generally focuses on a limited number of adult obesity determinants. We posited that more fully explaining Black-White female obesity inequities requires greater attention to multiple life course stages and obesity determinants including upstream and proximate determinants. Results from this study support this notion. Analysis of data from a national sample of Black and White women found that socioeconomic and social disadvantages such as living in disadvantaged neighborhoods and single parent households as adolescents and lower adult household incomes explain the majority of group differences in the obesity prevalence of Black and White U.S. women. Population health initiatives aimed at tackling racialized inequities in obesity will be most effective if they focus on systemic and structural determinants rather than individual level behavioral factors alone. Moreover, interventions targeting individuals earlier in the life course would help to alleviate Black-White obesity inequities among women in the U.S.

Introduction

Obesity is higher in the United States (U.S.) than in any other OECD country (OECD 2023). The most recent Centers for Disease Control and Prevention (CDC) estimates indicate that 40% of all U.S. adults are obese (Emmerich et al. 2024). This is a major population health concern because obesity is a leading cause of U.S. premature mortality and disability (Al Snih et al. 2007; Censin et al. 2019; National Academies of Sciences and Medicine 2021). As of 2016, excess body weight was associated with an average loss in U.S. life expectancy of approximately 2.4 years, with obesity contributing to over 1,000 excess deaths per day—on par with smoking (Ward et al. 2022).

As is evident with most U.S. health issues, the risk of obesity is inequitably distributed across subpopulations, with wide racialized inequities in obesity prevalence, especially among U.S. women. The age-adjusted obesity prevalence among non-Hispanic White (hereon “White”) women is 39.6%, and while this prevalence is high, more than half (57.9%) of non-Hispanic Black (hereon “Black”) women have obesity (Stierman et al. 2021). More generally, the obesity prevalence among Black women is higher than that of any other group defined by both sex and race or ethnicity, and when estimates of overweight and obesity are considered together, 79.6% of Black women are classified in these two weight categories (Office of Minority Health 2022). This places Black women at an undue risk of hypertension, cardiovascular disease (Go et al. 2013), diabetes (Krishnan et al. 2007), obesity-attributable mortality (Masters et al. 2013), and all-cause mortality (Boggs et al. 2011).

In a recent review, Frisco et. al. (2022) argued that a serious limitation in population health research on U.S. racial and ethnic inequities in obesity is the limited attention that has been given to Black women’s obesity vulnerability. This body of research is not nearly as extensive as the attention given to other groups including U.S. Hispanic adults, immigrants, and male children of immigrants. Frisco et. al. (2022) called for more research aimed at understanding factors that place Black women at such a high obesity risk relative to other groups, arguing that this will produce knowledge that can be used to implement actions to reduce Black women’s obesity risk.

To date, the limited research investigating factors explaining Black-White inequities in obesity among women have focused on point-in-time proximate behavioral factors (e.g., caloric intake and physical activity), socioeconomic status, and/or demographic indicators (Johnston and Lee 2011; Sen 2014; Siddiqi 2018). Findings from these studies left most of the Black-White inequities in body weight unexplained. This is unsurprising. Previous research has given limited attention to upstream determinants of health that have long been implicated as fundamental causes of health inequities (e.g., Phelan and Link 2015; Williams 1997; Williams et al. 2019). Moreover, factors that produce disparities accumulate over time and the life course as is evident by the fact that Black-White obesity inequities are not observed among girls until adolescence (Ogden et. al. 2018).

This study’s purpose is to expand knowledge about factors that explain Black-White obesity inequities among U.S. women by analyzing data from the National Longitudinal Study of Adolescent to Adult Health (Add Health). We investigate how different aspects of women’s adolescent and adult lives produce obesity inequities between Black and White women ages 33-44 in a nationally representative sample. A study strength is our ability to leverage longitudinal data to estimate how a wide variety of factors produce obesity disparities, including earlier life weight, adolescent and adult (dis)advantage, adult discrimination, depressive symptoms and health behaviors during adolescence and adulthood, and family experiences. We use logistic regression and Oaxaca-Blinder regression decomposition (logistic regression) techniques, which allow us to divide Black-White differences in obesity among women into parts that are explained by group differences in the distributions of the factors just cited (differences in observable characteristics) and unexplained by the observed factors (differences in the coefficients or effects). We expect our analysis to have greater explanatory power than analyses in previous research because we account for a wide array of predictors measured at multiple points in women’s lives.

Background

Public perceptions of obesity risk generally focus on individuals’ willpower and discipline with respect to eating and exercise (Saguy 2013). Yet much scholarship links macrolevel changes in human environments to population-level increases and inequities in obesity. For example, Nestle (2002) describes how the U.S. food environment has an overabundance of foods with minimal nutritional value due to the expansion of efficient food production and distribution systems that began after World War II (Popkin 2008). These structural forces changed individuals’ eating behaviors and U.S. diets in ways that had major implications for the high U.S. obesity prevalence. Individual-level changes in physical activity are also largely due to structural factors including the changing nature of work and leisure (Brownson et al. 2005).

These macrolevel changes do not equitably impact everyone in society. For example, Mirowsky and Ross (2015) argue that education increases the ability to navigate these health contexts. Others have extended their arguments by applying them to racial and ethnic inequities in obesity (e.g., Frisco et al. 2016) and arguments about systemic racism’s consequences for the acquisition of socioeconomic resources among racially marginalized groups, which in turn limits their access to nutritious food and health-benefitting environments (Hargrove 2018). As such, Black-White inequities in women’s obesity risk likely reflect gendered and racialized U.S. social structures and the way that Black and White women are situated within them across their lives. This makes it imperative to consider individual-level and structural factors that contribute to Black-White inequities in women’s obesity prevalence.

Evidence is clear that Black women in the U.S. face more social and economic disadvantages than White women (e.g., Essed 1991; Hargrove 2018). Black women have a greater likelihood of having disadvantaged backgrounds (Kravitz-Wirtz 2016). On average, they have lower odds of college degree attainment (U.S. Department of Labor 2022), lower earnings (Kochhar 2023), and a higher likelihood of living in poverty (Sun 2023). Black women are also disproportionately more likely to encounter institutional and interpersonal discrimination and racial and economic segregation (Geronimus et al. 2010). These are all factors that are associated with obesity. In addition to hindering access to health-benefitting resources, socioeconomic disadvantage also produces stress, which is associated with consumption of more energy-dense foods high in sugar and fat (O'Connor and Conner 2011; Torres and Nowson 2007), likely due to physiological mechanisms such as increased cortisol (Tomiyama et al. 2013; Torres and Nowson 2007) and using food to cope with stress (Geronimus et al. 2010).

Studies also directly implicate discrimination and racism in Black women’s high obesity risk and not simply due to the way that they produce socioeconomic disadvantage. Hunte (2011) and Hunte and Williams (2009) find that discrimination increases the risk of obesity among Black women and other minoritized groups defined by sex, race, and ethnicity. Discrimination has also been linked to excessive pregnancy weight gain among Black women, which is an important determinant of later life obesity (Reid et al. 2016). Vigilance against discrimination (Hicken et al. 2018) and unfair treatment in employment, the housing sector, and by police (Cozier et al. 2014) are all also positively associated with Black women’s obesity risk.

So is racial segregation. Black-White inequities in obesity among women are not evident in integrated census tracts (Bleich et al. 2010) and Black women’s obesity risk is greater in high-poverty, highly concentrated Black neighborhoods relative to other neighborhood contexts (Gailey and Bruckner 2019). This is likely due to environmental stressors and factors that hinder residents’ ability to exercise and access nutritious foods, including the preponderance of fastfood restaurants and limited recreation facilities and supermarkets in segregated neighborhoods (Phelan and Link 2015). Although studies do find that neighborhood-level racial isolation, social disorder, and community disadvantage are related to women’s obesity risk, they do not fully explain significant differences in Black and White women’s obesity risk (Chang et al. 2009; Robert and Reither 2004).

These differences appear to begin in childhood. Analysis of cross-sectional data suggests that significant weight inequities between White and Black girls first emerge among school-aged children ages 6-11 (Ogden et al. 2018). Black girls then face larger increases in weight than White girls as they age, and parents’ net wealth during adolescence does not protect Black girls from weight gain the same way that it does for White girls, most likely due to the racism-related stressors that Black girls face (Hargrove 2018).

The empirical research on other reasons for Black-White female weight disparities focus on proximate and socioeconomic determinants. For example, decomposition of Black-White inequities in obesity among women two decades ago focused on caloric intake and energy expenditures. Only caloric input significantly contributed to Black-White obesity inequities (Johnston and Lee 2011) in line with evidence about the importance of diet versus physical activity for body weight (Dugas et al. 2011; Pontzer et al. 2012). Nonetheless, caloric input explained only around 10% of Black-White female obesity inequities. Decomposition of cross-sectional data from Black and White women in Alabama and Mississippi was also unable to explain obesity inequities, with income, educational attainment, and health insurance explaining less than 10% of racialized weight inequities (Sen 2014).

This study extends research aimed at explaining Black-White inequities in obesity among women. Our analysis gives attention to structural, economic, social, and behavioral determinants of obesity during adulthood and adolescence. We consider obesity determinants during adolescence because it is a critical period for the development of health-related behaviors that have long-term implications for weight (Lee et al. 2013). For example, adolescent school (Lee et al. 2013) and neighborhood (Alvarado 2019; Burdette and Needham 2012; Lippert 2016) disadvantage are associated with adult obesity risk, with adolescent neighborhood disadvantage being especially consequential for Black individuals (Kravitz-Wirtz 2016) and Black young women, in particular (Nicholson and Browning 2012).

To our knowledge, our study is the first to decompose Black-White female obesity disparities into such a wide array of obesity determinants at different life course stages. We expect our approach to lead to a greater understanding of drivers of weight inequities between Black and White women, and as a result, factors that interventions can target to mitigate them.

Methods

Data and Sample

Data were drawn from Waves I, II, and V of Add Health, a longitudinal nationally representative study of 1994-95 7^th-12^th graders (ages 11-19). Wave I data were collected in 1994-95 from over 90,000 students across 145 middle, junior, and high schools, with 20,745 respondents selected for in-home surveys. Wave I also includes parent/guardian survey data. To date, four follow-up data waves have been collected from Wave I respondents in 1996 (Wave II), 2001-02 (Wave III), 2008-09 (Wave IV), and 2016-18 (Wave V), with survey response rates among eligible participants of 88.6% (Wave II), 77.4% (Wave III), 80.3% (Wave IV), and 71.8% (Wave V) (https://addhealth.cpc.unc.edu/documentation/study-design/). In addition to survey data, other types of participant data were collected across the Waves, including contextual (e.g., geolocation) and biomarker data.

Our sample is drawn from the 5,381 Wave V respondents (out of the total 12,300 Wave V respondents ages 33-44) who participated in in-home health examinations (i.e., the biomarker sample) that included measurement of individuals’ weight and height. We restrict the sample to 2,725 women who self-identified as White or Black. We excluded 64 pregnant women because their body weight does not capture normal body composition, and 63 additional women missing information needed to account for Add Health’s complex sampling design (Chen and Harris 2020). We do not exclude participants with missing values on analytic variables because we employ the Stata version 18.0 mi function (StataCorp, College Station, Texas) to impute missing data using “chained” equations (MICE) over 20 iterations. Our final analytic sample includes 2,598 White and Black women.

Variables

Our outcome is Wave V obesity (1=yes), based on respondents’ measured weight and height. From these data, we calculated BMI (weight (kg) / [height (m)]²) and obesity (BMI ≥ 30.0) using Centers for Disease Control and Prevention (CDC) guidelines.

Our primary independent variable is race. Women in the sample self-identified as “White” and “not Hispanic” (=0) or “Black, African American” and “not Hispanic” (=1).

The study includes numerous measures that may help to explain Black-White obesity inequities. These measures can be classified into categories indicating early life weight, (dis)advantage, adult discrimination experiences, depressive symptoms, health behaviors including proximate weight determinants, family life, and age.

We measure early life weight using respondents’ birthweight and adolescent obesity. Birthweight (continuous) was parent-reported in pounds and ounces, which we converted to grams. Adolescent obesity (1 = yes) is constructed from Wave II measured height and weight used to compute adolescent CDC age- and sex-specific BMI percentiles. The CDC classifies adolescents with a BMI percentile at or above the 95^th as obese. We did not use Wave I data to estimate adolescent obesity because height and weight are self-reported, not measured.

We measure (dis)advantage with individual-, family-, and neighborhood-level factors using data from adolescence (Wave I) and adulthood (Wave V). Adolescent measures include family socioeconomic status, neighborhood disadvantage, and neighborhood segregation. Adolescent family socioeconomic status (continuous) is a composite measure constructed by Belsky et al. (2018) using information from parents/guardians at Wave I on parental education, parental occupation, household income, and household receipt of public assistance. Belsky and colleagues used factor loadings from a principal component analysis to construct a composite index with Z-transformed values. Higher scores indicate higher family socioeconomic status.

Adolescent neighborhood disadvantage (continuous) was constructed by Belsky et al. (2019) using Census-tract-level data from the 1990 Decennial Census linked to Wave I participants’ addresses in 1994-1995. These data were used to compute census tract decile scores for each census tract for the following measures: proportion of female-headed households, individuals living below the poverty threshold, individuals receiving public assistance, adults with less than a high school education, and unemployed adults. Neighborhood disadvantage is the sum of decile scores across the five measures (Belsky et al. 2020) with higher scores indicating more disadvantaged neighborhoods.

Adolescent neighborhood segregation (continuous) was constructed by Add Health researchers using Wave I geolocation (census tract) information combined with data from the 1990 Decennial Census and the American Community Survey (Schwartz 2023). This neighborhood-level measure indicates racial residential segregation of Black Americans from other racial and ethnic groups as calculated by the G* statistic (see Schwartz 2023 for more details). Higher segregation scores indicate that a tract and its neighboring tracts have a higher proportion of Black residents than the average within their county.

Like previous research aiming to explain Black-White weight disparities among women, our models include measures of adult (Wave V) socioeconomic status. Our indicators include income, educational attainment, and employment status (1=unemployed). The Wave V household income measure is reported in discrete categories ranging from $5,000 or less to $200,000 or more. To create an adjusted household income measure, we calculated the midpoint of each category and then adjust for household size by dividing the midpoint by the square root of the participants’ household size. We then transform this measure by using the natural log to reduce positive skew (Assini-Meytin et al. 2022). Highest level of educational attainment indicates if women had “a high school degree or less (reference),”, “some college”, or “a college degree or more.”

Adulthood neighborhood disadvantage (continuous) at Wave V is a similar indicator to adolescent neighborhood disadvantage. It is constructed by Add Health researchers (Gaydosh et al. 2024) using data from American Community Survey (ACS) 5-year (2014-2018) estimates to create a census tract-level socioeconomic disadvantage score based on a factor analysis of tract proportion of the population ages 25 and older with less than high school diploma, proportion of female headed family households, proportion of the civilian population ages 16 and older currently unemployed, logged median household income, and the proportion of individuals living in poverty. Adulthood neighborhood segregation was also similarly constructed by Add Health researchers at Wave V as at Wave I (see above) using data from the 2010 Decennial Census and American Community Survey.

Along with (dis)advantage, we include measures of discrimination experiences at Wave V: whether women experienced unfair police treatment (1=yes), a major discriminatory experience (Williams et al. 1997), and frequency of interpersonal discrimination (range: 0 – 15). The interpersonal discrimination measure is based on five questions from a modified version of the Everyday Discrimination Scale (Williams et al. 1997). Respondents were asked: “In your day-to-day life, how often have any of the following things happened to you? 1) You are treated with less courtesy or respect than other people; 2) You receive poorer service than other people at restaurants or stores; 3) People act as if they think you are not smart; 4) People act as if they are afraid of you; and 5) You are threatened or harassed.” Responses include: “never”, “rarely”, “sometimes”, or “often,” which we code as “0,” “1,” “2,”and “3,” respectively, and then produce a final sum across the five questions ranging from 0 – 15 (Cronbach’s alpha=0.74) in line with previous research (e.g., Seng et al. 2012). Note that we do not include measures of adolescent discrimination because comparable measures to those in adulthood are not available.

Measures of adolescent and adult mental health and health behaviors include depressive symptoms, being a smoker (1=yes), binge drinking frequency, exercise frequency, and screen time hours. We also include adult fast food consumption frequency, which was unfortunately not available at Wave I. Adolescent depressive symptoms (range: 0 – 57) is based on a 19-item version of the Center for Epidemiological Studies (CES-D) instrument that asked how often in the past week respondents experienced symptoms of depression such as poor appetite, not being able to shake off the blues, not feeling as good as other people, and feeling depressed. Each question’s response options included “never or rarely=0”, “sometimes=1”, “a lot of the time=2”, or “most of the time or all of time=3”. We summed the responses to the 19 questions to create the composite measure (Cronbach’s alpha=0.88). Adult depressive symptoms is based on an abbreviated 5-item CES-D instrument asking respondents how often in the past week they felt “depressed,” “happy,” “sad,” “that life was not worth living,” and “that they could not shake off the blues.” Response options were coded the same as Wave I and we summed the responses to create a scale ranging from 0-15 (Cronbach’s alpha = 0.84). For both the adolescent and adult depressive symptoms measures, higher scores indicate more depressive symptoms.

Adolescent binge drinking frequency (range: 0-6) is based on a Wave I question that asked respondents how often in the past 12 months they drank 5 or more drinks in a row. Response options ranged from “every day or almost every day=0” to “never=6”. We reverse coded the measure so that higher scores indicate greater frequency of binge drinking. Adult binge drinking frequency (range: 0-6) is based on a similar Wave V question that asked respondents in the past 12 months how many days they drank 4 or more drinks in a row with response options ranging from “none=0” to “every day or almost every day=6”.

The exercise frequency measures were constructed from a series of Wave I and Wave V questions that asked about frequency of participation in various physical activities. Adolescent exercise frequency (range: 0-9) is based on three questions that asked about how many times in the past week respondents engaged in activities such as rollerblading and bicycling, exercising (e.g., jogging, walking, and karate), and playing an active sport such as basketball, soccer, football, or swimming. Response options for each of the Wave I questions include “not at all=0”, “1 or 2 times=1”, “3 or 4 times=2”, or “5 or more times=3”. We summed responses to the three questions to create the composite measure. Adult exercise frequency (range: 0-42) is based on 6 questions asking about a range of activities individuals participated in the past 7 days (e.g., cycling, running, weight training, organized sports). Response options for each question range from “0” to “7” times, which allowed us to create a summary score ranging from 0 to 42.

Adolescent screen time hours (range: 0-80) is a summed measure of three Wave I questions that asked respondents how many hours in the past week they spent watching television and videos, playing video games, or playing computer games. Following other studies, we truncated this measure at 80 (e.g., Boone et al. 2007). Adult screen time hours (range: 0-80) is based on a Wave V question asking the number of times in the last week individuals watched “television, movies or videos, including DVDs or music videos”, which we also truncated at 80. Finally, adult fast food consumption frequency (range: 0-28) is based on a Wave V question asking the total number of times in a week individuals eat food from fast food outlets.

Indicators of family life in the analysis include the composite Add Health adolescent household structure measure (Harris 1999), which we use to designate whether individuals lived with “two biological parents”, “two parents,” “a single parent,” or “other.” We also include measures of adult Wave V marital status (1 = unmarried), life course number of pregnancies (range: 0-20), and whether a respondent has children living in the household (1 = yes) at Wave V.

Our final analytic variable is Wave V age, which ranges from 33-44. It is included because on average, individuals generally gain weight as they age (e.g., Zheng et al. 2017).

Methodological Approach

We first present descriptive statistics for the sample and for White and Black women separately in Table 1 to show group differences in obesity and factors that may explain them. We then show estimates from two weighted logistic regression models in Table 2 that include only race (Model 1) and then all other study variables (Model 2). We then use Oaxaca-Blinder twofold decomposition techniques for nonlinear (logistic) models by using the logit extension of Jann’s (2008) original Oaxaca command with the normalize function for categorical variables in Stata version 18.0 (StataCorp, College Station, Texas). The results from our decomposition analysis are shown in Figures 2 and 3.

Table 1.

Descriptive statistics for the total sample and by race

	Total sample	White women	Black women
	N = 2,598	N=1,900	N=698
	Mean (SE) or %	Mean (SE) or %	Mean (SE) or %
Wave V Adult obesity (BMI ≥ 30.0)	48.5	43.8	67.0***
Early Life Weight
Birthweight (grams) (range: 1814.40 – 5216.40)	3300.86 (16.60)	3330.59 (17.87)	3183.85 (30.65)***
Adolescent obesity (≥ 95th BMI percentile)	13.5	11.1	22.7***
(Dis)advantage
Adolescent family socioeconomic status (range: −4.66 - 3.39)	0.10 (0.07)	0.29 (0.07)	−0.66 (0.12)***
Adolescent neighborhood disadvantage (range: 5 - 50)	25.04 (0.97)	22.20 (0.99)	36.21 (0.98)***
Adolescent neighborhood segregation (range: −3.34 - 11.85)	−0.13 (0.17)	−0.68 (0.11)	2.02 (0.34)***
Adult adjusted household income (range: 6.85 - 12.43)	10.37 (0.04)	10.50 (0.05)	9.87 (0.08)***
Adult highest level of educational attainment
HS degree or less	13.8	13.2	16.0
Some college	42.2	41.1	46.8
College or more	44.0	45.7	37.2*
Adult unemployed (1=yes)	23.2	23.3	22.5
Adult neighborhood disadvantage (range: −1.79 - 5.13)	−0.12 (0.06)	−0.29 (0.06)	0.57 (0.08)***
Adult neighborhood segregation (range: −3.78 - 13.18)	−0.19 (0.10)	−0.58 (0.06)	1.38 (0.16)***
Adult Discrimination
Frequency of discrimination (range: 0 - 15)	3.35 (0.09)	3.16 (0.10)	4.10 (0.17)***
Unfair police treatment (1=yes)	12.6	10.0	22.8***
Depressive Symptoms
Adolescent depressive symptoms (range: 0 - 50)	11.57 (0.25)	11.21 (0.29)	12.96 (0.53)**
Adult depressive symptoms (range: 0 - 15)	2.52 (0.11)	2.47 (0.12)	2.75 (0.20)
Health Behaviors
Adolescent smoker (1=yes)	22.2	25.7	8 7***
Adolescent binge drinking frequency (range: 0 - 6)	0.53 (0.04)	0.59 (0.05)	0.30 (0.06)***
Adolescent exercise frequency (range: 0 - 9)	3.27 (0.08)	3.40 (0.09)	2.78 (0.12)***
Adolescent screen time hours (range: 0 - 80)	19.25 (0.73)	17.30 (0.70)	26.90 (1.66)***
Adult smoker (1=yes)	22.8 33	23.1	21.7
Adult binge drinking frequency (range: 0 - 6)	0.90 (0.03)	0.95 (0.04)	0.69 (0.07)**
Adult exercise frequency (range: 0 - 37)	6.39 (0.18)	6.57 (0.20)	5.65 (0.36)*
Adult fast food consumption frequency (range: 0 - 28)	1.83 (0.07)	1.64 (0.08)	2.56 (0.15)***
Adult screen time hours (range: 0 - 80)	12.94 (0.38)	12.47 (0.41)	14.80 (1.07)*
Family Life
Adolescent household structure
Two biological parents	55.8	62.6	29.0***
Two parents	15.8	16.6	12.8
Single parent	23.0	17.2	45.6***
Other	5.4	3.6	12.6***
Adult unmarried (1=yes)	42.5	36.1	67.9***
Life course number of pregnancies (range: 0 - 20)	2.26 (0.05)	2.18 (0.05)	2.57 (0.13)**
Adult has children living in household (1=yes)	74.8	75.1	73.6
Adult Age (range: 33 - 44)	37.76 (0.13)	37.70 (0.15)	37.98 (0.22)

^***p<0.001

^**p<0.01

^*p<0.05 different from White women based on Chi-Squared or t-test

Source: National Longitudinal Study of Adolescent to Adult Health, Waves I (1994-5), II (1996), and V (2016-2018)

Notes: All variables measured in adolescence were constructed from Wave I data collected when respondents were 7^th-12^th graders. All variables measured in adulthood were constructed from Wave V data collected when respondents were ages 33-44. The only exception is adolescent obesity, which was constructed using Wave II data; weight and height were measured at Wave II and self-reported during Wave I interviews.

Table 2.

Estimates (odds ratios) from logistic regression models of Wave V (2016-2018) adult obesity (BMI ≥ 30.0) among Black and White women ages 33-44 who participated in the National Longitudinal Study of Adolescent to Adult Health (N=2,598)

	Model 1		Model 2
	OR	95% CI	OR	95% CI
Black (ref=Non-Hispanic White)	2.61*** (1.93, 3.53)	1.20	(0.84, 1.73)
Early Life Weight
Birthweight			1.00	(1.00, 1.00)
Adolescent obesity (≥ 95th BMI percentile)			14.88***	! (7.89, 28.07)
(Dis)advantage
Adolescent family socioeconomic status			0.96	(0.86, 1.08)
Adolescent neighborhood disadvantage			1.01	(1.00, 1.03)
Adolescent neighborhood segregation			0.98	(0.90, 1.07)
Adult adjusted household income			0.78**	(0.66, 0.93)
Adult highest level of educational attainment (ref=HS degree or less)
Some college			0.84	(0.58, 1.22)
College or more			0.58*	(0.39, 0.88)
Adult unemployed (ref=employed)			0.87	(0.60, 1.26)
Adult neighborhood disadvantage			1.16	(0.97, 1.38)
Adult neighborhood segregation			1.01	(0.95, 1.08)
Adult Discrimination Experiences
Frequency of discrimination			1.02	(0.97, 1.08)
Unfair police treatment (ref=no)			0.99	(0.64, 1.51)
Depressive Symptoms
Adolescent depressive symptoms			0.99	(0.97, 1.00)
Adult depressive symptoms			0.98	(0.93, 1.03)
Health Behaviors
Adolescent smoker (ref=no)			1.02	(0.75, 1.40)
Adolescent binge drinking			0.95	(0.85, 1.07)
Adolescent exercise frequency			1.02	(0.96, 1.08)
Adolescent screen time hours			1.00	(0.99, 1.01)
Adult smoker (ref=no)			0.76	(0.53, 1.09)
Adult binge drinking frequency			1.03	(0.94, 1.12)
Adult exercise frequency			0.96**	(0.94, 0.99)
Adult fast food consumption frequency			1 16***	(1.08, 1.24)
Adult screen time hours			1.01*	(1.00, 1.02)
Family Life
Adolescent household structure (ref=two bio parents)
Two parents			0.82	(0.60, 1.13)
Single parent			1.34	(0.97, 1.86)
Other			1.05	(0.64, 1.74)
Adult unmarried (ref=married)			0.84	(0.64, 1.07)
Life course number of pregnancies			1.00	(0.93, 1.08)
Adult has children living in household			0.75	(0.55, 1.03)
Adult Age			1.02	(0.96, 1.09)

^***p<0.001

^**p<0.01

^*p<0.05

Notes: All variables measured in adolescence were constructed from Wave I data collected when respondents were 7^th-12^th graders. All variables measured in adulthood were constructed from Wave V data collected when respondents were ages 33-44. The only exception is adolescent obesity, which was constructed using Wave II data; weight and height were measured at Wave II and self-reported during Wave I interviews.

Figure 2.

Graphical depiction of results from an Oaxaca-Blinder decomposition (logistic regression) of Wave V (2016-2018) adult obesity (BMI ≥ 30.0) among Black and White women ages 33-44 who participated in the National Longitudinal Study of Adolescent to Adult Health (N=2,598)

Note: Each of the grouped ‘explained’ categories in the Figure include the following:

Early Life Weight: birthweight and adolescent obesity

(Dis)advantage: adolescent family socioeconomic status, adolescent neighborhood disadvantage, adolescent neighborhood segregation, adult adjusted income, adult highest level of educational attainment, adult unemployed, adult neighborhood disadvantage, and adult neighborhood segregation

Adult Discrimination Experiences: frequency of discrimination and unfair police treatment

Depressive Symptoms: adolescent depressive symptoms and adult depressive symptoms

Health Behaviors: Adolescent smoker, adolescent binge drinking frequency, adolescent exercise frequency, adolescent screen time hours, adult smoker, adult binge drinking frequency, adult exercise frequency, adult fast food consumption frequency, and adult screen time hours

Family Life: Adolescent household structure, adult unmarried, life course number of pregnancies, and adult has kids living in the household

***p<0.001 **p<0.01 *p<0.05

Figure 3.

Graphical depiction of results from Oaxaca-Blinder decomposition (logistic regression) of Wave V (2016-2018) adult obesity (BMI ≥ 30.0) among Black and White women ages 33-44 who participated in the National Longitudinal Study of Adolescent to Adult Health (N=2,598)

***p<0.001 **p<0.01 *p<0.05

Notes: All variables measured in adolescence were constructed from Wave I data collected when respondents were 7^th-12^th graders. All variables measured in adulthood were constructed from Wave V data collected when respondents were ages 33-44. The only exception is adolescent obesity, which was constructed using Wave II data; weight and height were measured at Wave II and self-reported during Wave I interviews.

Together results from logistic regression and decomposition provide two important but different pieces of information. Logistic regression results show differences in the odds of obesity between Black and White women net of model covariates whereas decomposition results show the group-level difference in the obesity proportions between Black and White women partitioned into parts that are explained by group compositional (or distribution) differences in the obesity predictors and unexplained by group compositional differences (i.e., structural effects). In other words, in addition to the explained portion, the decomposition also produces a residual that cannot be explained by group differences in the levels of the measured factors and derives from differences in the coefficients in the group-specific logistic regression models (i.e., differences in behavioral responses).

The Oaxaca-Blinder decomposition technique employs a counterfactual approach by first estimating separate logistic regression models predicting obesity for each group (i.e., Black and White women) and then decomposing the group-level obesity difference into the explained and unexplained parts by replacing the comparison group’s coefficients with the reference group’s coefficients. In our analysis, White women are the reference group, and Black women are the comparison group.

Figure 1 illustrates the nonlinear Oaxaca-Blinder decomposition of obesity ($y$) between Black (B) and White (W) women based on the nonlinear decomposition equation below. Nonlinear decomposition models do not follow the same logic as linear decomposition models because the average of the dependent variable ($\overline{Y}$) does not necessarily equal the function of the row vector of the average values of the independent variables and vector coefficient estimates ($F(\overline{X}\widehat{\beta })$ (Fairlie 2005). Researchers such as Farlie (2005) have extended the linear Oaxaca-Blinder decomposition model to nonlinear models, including logistic regression. Based on Fairlie’s logic, for our purposes, the nonlinear decomposition can be expressed as:

$${\overline{Y}}^B-{\overline{Y}}^W=\underset{ExplainedPart}{\underbrace{\left[\sum _{i=1}^{N^B}\frac{F({\widehat{\beta }}^W{X}_i^B)}{N^B}-\sum _{i=1}^{N^W}\frac{F({\widehat{\beta }}^W{X}_i^W)}{N^W}\right]}}+\underset{UnexplainedPart}{\underbrace{\left[\sum _{i=1}^{N^B}\frac{F({\widehat{\beta }}^B{X}_i^B)}{N^B}-\sum _{i=1}^{N^B}\frac{F({\widehat{\beta }}^W{X}_i^B)}{N^B}\right]}}$$

where ${\overline{Y}}^B$ is the average probability of obesity for Black women and ${\overline{Y}}^W$ is the average probability of obesity for White women. We specifically subtract White women’s probability from Black women’s probability because we anticipate Black women will have a higher average probability of obesity than White women. The first term on the left side of the equation is the explained part or composition component and the second term is the unexplained part or structural component of the decomposition. In our decomposition, we use White women’s coefficients as the reference coefficients¹. Specifically, the first term denotes the part of the gap in obesity between Black and White women that is due to group differences in the distributions of the independent variables ($X$) and the second term denotes the part due to differences in the coefficients or group processes.

Figure 1.

Graphical representation of the nonlinear Oaxaca decomposition of obesity (y) between Black (B) and White (W) women

The composition and structural components of the decomposition are illustrated in Figure 1, where $f(x^W)$ is the distribution of $x$ for White women and $f(x^B)$ is the distribution of $x$ for Black women. Point A $(F({\beta }^W{X}_i^W))$ is the proportion of obesity among White women and Point C $(F({\beta }^B{X}_i^B))$ is the proportion of obesity among Black women. These estimates are obtained by averaging the predicted probability of being obese for all White women across their entire distribution of $x(f(x^W))$ and for all Black women across their distribution of $x(f(x^B))$. Point B in the Figure is the predicted proportion of obesity among Black women if they had the same responses to behavioral and environmental risk factors as White women but retained their own characteristics. It is obtained by evaluating the predicted probability using White women’s logistic regression coefficients but Black women’s distribution on $x$. Therefore, the difference between points A and B is the difference in obesity prevalence attributable to compositional differences, while the difference between points B and C is due to unexplained or “structural” differences (i.e., group differences in their responses or vulnerability to risk factors).

Results

Descriptive statistics shown in Table 1 indicate that 48.5% of women are classified as obese. This estimate masks statistically significant differences between Black and White women. Roughly two thirds of Black women are classified as obese compared to 43.8% of White women.

There are also inequities in early life weight. Black women have significantly lower birthweights than White women, but weight patterns reverse by adolescence. Only 11.1% of White women had adolescent obesity compared to 22.7% of Black women.

There are also statistically significant racial inequities in (dis)advantage. During adolescence, Black women’s family socioeconomic status is significantly lower, and they live in more disadvantaged and racially segregated neighborhoods than White women. For example, a G* statistic score of 1.96 indicates significant racial segregation, or that a census tract and its neighboring tracts have a higher proportion of Black residents than the average in the county by approximately two standard deviations above the mean (Schwartz 2023). Black women’s average neighborhood segregation score in adolescence is 2.02 compared to White women’s score of −0.68. As adults, the average adjusted household income of Black women is also significantly lower than White women and a larger proportion of White women have a college degree relative to Black women (45.7% vs. 37.2%), but there are not significant racial differences in unemployment. As was the case in adolescence, Black women also live in more disadvantaged and more segregated neighborhoods than White women as adults.

Black women also report more discrimination. Nearly a quarter of Black women report unfair police treatment compared to only 10.0% of White women and the average frequency of interpersonal discrimination reported by Black women is significantly higher (4.10 vs. 3.16).

With respect to depressive symptoms and health behaviors, Black women have slightly higher but significantly different average depressive symptoms scores in adolescence relative to White women, but adult depressive symptoms are substantively and statistically similar. Adolescent smoking and binge drinking are significantly higher among White versus Black women, but so is exercise frequency. In adulthood, racial differences in smoking are no longer statistically significant but differences in binge drinking and exercise remain. Black women also have significantly higher screen time in adolescence and adulthood and more adult fast-food consumption than White women.

Finally, there are significant differences in the family experiences of Black and White women. Only 29.0% of Black women lived with both biological parents as adolescents compared to 62.6% of White women and as adults, significantly more Black women are unmarried, and their average number of pregnancies is slightly higher.

We now turn to estimates from logistic regression models predicting the odds of adult obesity in Table 2². The estimate from Model 1 reaffirms the statistically significant racial inequity in obesity among women. Black women’s estimated odds of having obesity are 2.61 times higher than White women’s odds. When study covariates are included in Model 2, the large, statistically significant difference in the odds of obesity between Black and White women is reduced by more than half and it is no longer statistically significant (OR=1.20, 95% CI: 0.84-1.73).

In Model 2, adolescent obesity is significantly and positively associated with the odds of obesity. This is unsurprising given the strong correlation between adolescent and adult weight. In addition, adult household income is negatively associated with the odds of obesity, as is earning a college degree relative to having a high school degree or less. Adult exercise frequency is also inversely related to women’s odds of obesity while fast food frequency is positively associated with women’s odds of obesity. Each additional hour of adult screen time also increases the odds of obesity by 1% (OR=1.01, 95% CI: 1.00-1.02).

We now turn to estimates from the decomposition analysis aimed at explaining the obesity inequity between Black and White women in Figure 2. The top panel of Figure 2 graphically displays the absolute ‘total difference’ in obesity between Black and White women, and amount of this total difference that is ‘explained’ and ‘unexplained’. The numerical estimates from this panel and the second panel of Figure 2 are shown in Appendix A1. Of the 0.232 difference in the proportions of Black and White women with obesity, 95.7% (0.222) is explained by the variables in the decomposition analysis (see Row 4 in Appendix A1).

The bottom panel of Figure 2 shows the absolute contributions of different groups of covariates representing key study conceptual constructs that explain the total difference in the proportion of Black and White women with obesity. For parsimony, we only graphically depict the ‘explained’ difference because of the minimal (0.010) and insignificant contribution of the ‘unexplained’ difference.

Figure 2 indicates that (dis)advantage during adolescence and adulthood plays the largest role in explaining the obesity inequity between Black and White women. Together, indicators of (dis)advantage explain slightly more than half of the Black-White female obesity difference (or 0.127 of the 0.222 explained difference). As a group, disadvantage indicators make a stronger contribution than even early life weight indicators, which also make significant contributions to the Black-White obesity difference along with health behaviors. Of the 0.222 explained difference, early life weight and health behaviors contribute 0.053 (22.9%) and 0.035 (15.1%), respectively (see Panel 2 in Appendix Table A1).

Figure 3 presents more fine-grained detail about how individual covariates in the decomposition analysis contribute to the ‘explained’ difference in obesity between Black and White women. Not surprisingly given the strong connection between earlier and later life weight, adolescent obesity is the individual covariate that makes the largest contribution. Adolescent obesity explains 23.7% (0.055/0.232=0.237 x 100) of the difference. Adolescent neighborhood disadvantage, adult household income, and adult neighborhood disadvantage also make important contributions to the disparity in obesity between Black and White women, explaining 21.6% (0.050/0.232=0.216 x 100), 15.1% (0.035/0.232=0.151 x 100), and 11.6% (0.027/0.232=0.116 x 100) of the difference between Black and White women, respectively. Note, though, that only the two former measures contributed to the difference in statistically significant ways. Together, these two measures alone explain slightly more than one-third of the difference in the proportion of Black and White women who have obesity. Adult fast-food consumption and living in a single parent household as an adolescent were the other two measures in the analysis that made large and statistically significant contributions to the Black-White obesity disparity among women.

Supplementary Analyses

We produced supplementary analyses to test the robustness of study results. We decomposed Black-White female obesity inequities omitting adolescent obesity due to how highly predictive adolescent obesity is of adult obesity. This was done to ensure that we were not underestimating the estimated effects of other weight determinants in the study. Results from this decomposition analysis are shown in Appendix Table A2, which indicates that we are still able to explain 88.4% of group inequities in obesity between Black and White women. (Dis)advantage and health behaviors continued to be the primary drivers of Black-White female obesity inequities. (Dis)advantage explained 68.1% of the difference and health behaviors explained 17.7% of the difference. More detailed estimates about individual measures in the decomposition (available upon request) show that adolescent neighborhood disadvantage, adulthood income, and adulthood fast-food consumption continue to contribute significantly to the explained proportion of the Black-White female obesity inequity.

We also produced OLS regression and Oaxaca-Blinder decomposition analyses of Black-White female inequities in BMI and waist-to-height ratio (WtH). Results from these analyses are shown in Appendix Tables A3 and A4. Panel 1 in Table A3 indicates that there is a 4.788 difference in the BMI of Black and White women. To put this difference into context, this represents a 28-pound difference for 2 women of average height (5 feet, 4 inches) in the U.S. The decomposition analysis explains 86.4% of the estimated BMI inequity between Black and White women. As was the case when we decomposed group differences in obesity, early life weight, disadvantage, and health behaviors significantly contribute to group differences in obesity between Black and White women. Oaxaca-Blinder decomposition (OLS regression) of adult WtH produce substantively similar results (see Table A4). Overall, these results confirm that the main factors that contribute to weight inequities between Black and White women are similar across different measures.

Discussion

In the U.S., Black women have the highest obesity prevalence of any racial, ethnic, and gender group (Office of Minority Health 2022), which contributes to a heightened risk of chronic conditions (Go et al. 2013; Krishnan et al. 2007) and reduced life expectancy (Boggs et al. 2011; Masters et al. 2013). This has motivated a small body of research aimed at explaining Black-White weight inequities among U.S. women. By analyzing cross-sectional data, these studies accounted for around 10% of the total estimated disparity in weight between Black and White women (Johnston and Lee 2011; Sen 2014; Siddiqi 2018). We posited that this was because these studies included a limited number of potential explanatory measures assessed at a single point in time.

In response, we analyzed longitudinal, nationally representative data collected during adolescence and adulthood to determine whether we could more fully explain obesity inequities between Black-White women with measures of structural, socioeconomic, social, and behavioral obesity determinants. Our approach aims to account for upstream and multiple determinants of body weight that accumulate over time to shape women’s risk of obesity and racialized obesity inequities between Black and White women.

Descriptive analysis indicated that Black and White women significantly differed on nearly every obesity determinant included in the study. Black women were significantly more likely than White women to have adolescent obesity despite their lower birthweight. As a group, Black women were also more disadvantaged than White women. Black women lived in more disadvantaged and more segregated neighborhoods as adolescents and adults than White women and Black women’s adolescent socioeconomic status, adult household income, and educational attainment was lower. Relative to White women, Black women were also significantly more likely to report interpersonal discrimination and unfair police treatment, and Black women consumed more fast food, exercised less frequently, and had higher levels of screen time in adolescence and adulthood. Black women were also more likely to live with single parents during adolescence, and they reported more pregnancies than White women.

These group differences were consequential for racial inequities in obesity. The decomposition analysis that we produced explained all but roughly 5% of the difference in the proportion of Black and White women in the study who had obesity. (Dis)advantage in adolescence and adulthood explained slightly more than half of the Black-White obesity inequity among women. When we did not include adolescent obesity—a powerful predictor of adult obesity—in supplementary analyses, adolescent and adult (dis)advantage accounted for roughly two-thirds of the Black-White obesity disparity among women. The two (dis)advantage indicators that significantly contributed to inequities individually were adolescent neighborhood disadvantage, which contributed to one-fifth of the overall obesity inequity between Black and White women, and adult household income, which explained 15% of group inequities.

In line with previous research (Johnston and Lee 2011), adult proximate obesity determinants also contributed to obesity inequities, with adolescent and adult health behaviors explaining 15% of Black-White female obesity inequities and fast-food consumption being a significant driver of this disparity on its own. This is consistent with previous research indicating that consumption is a more important determinant of weight than physical activity (Dugas et al. 2011; Pontzer et al. 2012).

We expected to find that neighborhood segregation (which we grouped with (dis)advantage measures due to the salience of structural racism for social disadvantage) and adult discrimination would play significant roles in explaining Black-White obesity inequities among U.S. women. Results did not support this expectation. Adolescent neighborhood segregation, adult neighborhood segregation, and the two study measures of adult exposure to discriminatory experiences were not independently associated with women’s odds of obesity net of other covariates nor did they contribute to group disparities in a statistically significant manner. It is possible that this is due to measurement limitations or due to these measures mattering indirectly through the ways that neighborhood segregation and discrimination shape disadvantage and health behaviors. Future research should explore these possibilities.

Despite this unexpected finding, study results provide three key takeaway points about factors that produce group inequities in obesity between Black and White U.S. women. First, studies that focus on single point-in-time factors that shape body weight in cross-sectional analyses will almost certainly have limited power to explain group inequities. In some ways, this is intuitive. Weight gain tends to be gradual, it generally occurs across the life course, and early life weight is one of the best predictors of later life weight (e.g., Guo et al. 2002; Wang et al. 2008). Indeed, in this study, adolescent obesity explains nearly a quarter of group inequities in obesity between Black and White women. Adolescent neighborhood disadvantage also played a significant role in the development of adult group obesity inequities between Black and White women as did living in a single parent household during adolescence. Altogether, differential exposures to adolescent neighborhood disadvantage and single parent households accounted for 32% of adult disparities in obesity.

These findings about adolescent neighborhoods and families suggest that they could be powerful contexts that shape factors such as stress and weight-related customs and norms in ways that have long-term implications for adult weight inequities between Black and White women. As such, failing to account for earlier life contexts will lead to limited understanding of why disparities emerge over time. This notion is consistent with evidence that eating behaviors (e.g., Nicklaus et. al. 2005) and patterns of physical activity are generally developed well before adulthood (e.g., Hirvensalo and Lintunen 2011), and that stress in adolescence has long-term implications for multiple cardiometabolic outcomes including fat distribution and obesity (Guo et. al. 2024). Our longitudinal analysis focused on adolescence and adulthood due to practical reasons (i.e., when data were collected) and evidence that adolescence is a critical life course stage for the formation of lifelong health- and lifestyle-related behaviors that are shaped by adolescents’ social environments (Lee et al. 2013). Future research should account for the importance of experiences during earlier life course stages as well.

A second takeaway of the current study is that racial inequities in obesity between Black and White women in large part reflect socioeconomic inequities and (dis)advantage and how these factors shape women’s risk of obesity. Adolescent neighborhood disadvantage, adult household income, and adolescent residence in single parent households all contribute to obesity inequities in statistically significant ways independently and as part of a larger constellation of (dis)advantages that contribute to more than half of the group differences in obesity between Black and White women in this study.

Our final study takeaway is that individual-level health behaviors do matter for group inequities in obesity between Black and White women, but when considered as part of a broader constellation of social and structural determinants of weight, they have less explanatory power than social and socioeconomic (dis)advantages. The entire array of health behaviors we measure in adolescence and adulthood explains only 15% of group obesity inequities between Black and White women. Even when we do not account for adolescent obesity, all health behaviors account for 17% of group obesity inequities. Part of the limited explanatory power of health behaviors is almost certainly measurement error in assessing eating, drinking, smoking, and physical activity in survey data. Yet we contend that even without measurement error, other social determinants—and socioeconomic determinants, in particular—are likely more important because they are upstream determinants of a multitude of factors in women’s lives that shape weight, obesity, and racial inequities in these outcomes.

Study results and our conclusions about them must be considered in light of limitations. First, we cannot speak to causality. Second, although not the purpose of our study, we do not uncover the reasons why our explanatory factors contribute to obesity inequities. For example, we only speculate about why adolescent neighborhood disadvantage contributes to Black-White obesity inequities. Along this vein, because we do not have information about respondents’ neighborhoods prior to Wave I, we cannot ascertain if (or how) neighborhood exposures prior to Wave I (i.e., in childhood and earlier adolescence) contribute to weight and obesity inequities. We also do not consider the potential importance of residential (im)mobility throughout the young adult life course. However, we included a comparable Wave IV (when respondents were ages 24-32) measure of neighborhood disadvantage in supplementary analyses not shown but available upon request. Including this Wave IV measure did not change findings, and it did not make a significant or substantial contribution to the inequities we studied. These findings further support the salience of adolescent neighborhood disadvantage in explaining the obesity inequity between Black and White women. Future research should explore the precise mechanisms through which adolescent (and earlier life course) neighborhood disadvantage contributes to body weight disparities and the potential role of residential (im)mobility.

We note two additional limitations before moving to study conclusions. Our measures of racism and discrimination are limited. For example, we do not have measures of vigilance against discrimination (Hicken et al. 2018), cultural racism, vicarious discrimination, and other macro-level measures of structural racism (Williams et al. 2019) that are associated with obesity (Bell et al. 2019). In addition, some Add health measures available in adulthood were not assessed in Wave I, including discrimination experiences and fast-food consumption, which means that we could underestimate the importance of earlier life experiences for adult obesity inequities between Black and White women.

Despite limitations, our study extends knowledge about Black-White inequities in obesity among women by showing that these disparities are best understood within a broad context that accounts for multiple life course stages and upstream and downstream obesity determinants. Indeed, socioeconomic and social disadvantages such as living in disadvantaged neighborhoods and single parent households as adolescents and lower adult household incomes explain the majority of the group difference in the obesity prevalence of Black and White U.S. women. Population health initiatives aimed at tackling racialized inequities in obesity will be most effective if they focus on inequities in systemic and structural determinants rather than individual level behavioral factors alone. Moreover, interventions targeting individuals earlier in the life course would likely go a long way in alleviating adult racialized obesity inequities.