The Role of Residential Segregation in Explaining Racial Gaps in Childhood and Adolescent Obesity

Race-Ethnic Gaps in Obesity

Although many factors contribute to the obesity epidemic, many researchers agree that obesity is a social disease, meaning that it mostly stems from larger social issues such as perpetual social inequalities based on race-ethnicity, socio-economic status, gender, and so forth (Williams & Collins, 1995; Wyatt et al., 2003). Race is one of the major bases of division in American life and has been a critical factor in the social and economic structure of American society from its pre-colonial beginnings to the present (Bonilla-Silva, 2006; White & Borrell, 2011). Furthermore, in the United States race has traditionally been a more important determinant of individual life chances than socio-economic status (Saperstein, 2012). Race is an antecedent of socio-economic status because institutionalized racism has blocked upward mobility for members of minority groups (Feagin, 2014). Given the history of racism in the United States and the ongoing reality of racial discrimination, it is not surprising that throughout U.S. history racial disparities in health have been pervasive (Subramanian, Acevedo-Garcia, & Osypuk, 2005; Williams & Collins, 1995). These disparities are a product of the race privilege according to which Whites have always been at the top, Blacks at the bottom, and other groups in between. Even now, after years of policies that promoted equality of opportunity, by every measure non-Hispanic White Americans enjoy better health that their Black co-nationals (Williams & Sternthal, 2010).

Traditionally, racial segregation has been a key mechanism by which racial inequality has been created and reinforced (Williams & Sternthal, 2010). Some researchers (e.g., White & Borrell, 2011; Williams & Collins, 2001) argue that racial disparities in health reflect, in part, the legacy of racial segregation that limits access to quality care for minority groups. Moreover, earlier studies have noted that segregated communities are unable to meet their residents’ dietary needs (Larson, Story, & Nelson, 2009; Story, Kaphingst, Robinson-O’Brien, & Glanz, 2008) because of a high concentration of fast food outlets (Morland & Evenson, 2009; M. C. Wang, Kim, Gonzales, MacLeod, & Winkleby, 2007) and a lack of community safety for outdoor physical activities (Kipke et al., 2007; Smith et al., 2011). Although there is limited literature on this topic (e.g., Larson et al., 2009; Subramanian et al., 2005; Williams & Collins, 2001), the relationship between racial segregation and childhood obesity has drawn little attention. To the best of our knowledge, there is only one study that explicitly tested the hypothesis whether and how that racial residential segregation contributes to obesity epidemic. Using Behavioral Risk Factor Surveillance System (BFRSS) data, Chang (2006) conducted multilevel analyses to compare weight status of non-Hispanic Blacks and Whites. Using racial isolation index as a neighborhood-level factor, she found that among non-Hispanic Blacks, higher racial isolation is positively associated with both a higher body mass index (BMI), after adjusting for multiple covariates, including measures of individual socio-economic status. It is also worth mentioning a rather similar study by Wickrama, Wickrama, and Bryant (2006) which, using the same dataset (BFRSS), examined the influence of community poverty on adolescent weight status. The authors found that adolescents living in poor communities were more likely to be obese. Building on these studies, we also use multilevel regression whereby the socio-economic effects at the individual level are nested at the neighborhood level. This technique can give us a better understanding of the relationship between racial segregation as a neighborhood-level factor and obesity as an individual-level outcome. However, the present study differs from Chang’s (2006) study in that our focus is not on adults, but on children and adolescents. Moreover, in contrast to Wickrama et al. (2006), we focus on residential segregation and not on community poverty, and we cover several obesity-related health outcomes of children as well as adolescents. Furthermore, unlike the two aforementioned studies, our study is longitudinal which will enable identifying causal factors that exert their influence over a prolonged period of time.

Other Risk Factors for Obesity

Although the favored model of risk for obesity is social, genetic factors have also been cited as those related to obesity in children and adolescents (Reilly et al., 2005; Sørensen, Price, Stunkard, & Schulsinger, 1989; Strauss & Knight, 1999). For example, Davis, McGonagle, Schoeni, and Stafford (2008) showed that parents’ and grandparents’ weight status also plays an important role among the causes of obesity. Another group of risk factors that has received prominence in the discussions of interventions to curb diabetes epidemic is related to socio-economic status (e.g., Bae et al., 2014; Levine, 2011; Zhang & Wang, 2004). Structural factors, such as parental income and education, are known to determine level of access to sedentary and fast food lifestyles, with decreased physical activity opportunities and increased availability of high-fat energy-dense foods (Biro & Wien, 2010; Chang, 2006; Gordon-Larsen, Harris, Ward, & Popkin, 2003; Reilly et al., 2005). More generally, it has been suggested that family socio-economic status is inversely related to prevalence of being overweight (Levine, 2011; Zhang & Wang, 2004). Apart from socio-economic status, family structure and size can affect the odds of overweight and obesity in children and adolescents (Chen & Escarce, 2010; Gable & Lutz, 2000). Generally, it has been documented that the presence or absence of a parent, as well as number of siblings may affect weight status of children (Chen & Escarce, 2010). Particularly, a strong relationship between single-parent status and excess weight in children has been suggested by Strauss and Knight (1999). Single-parent households and large households have a tendency to favor the consumption of prepared food items, which tend to be high in fat and sodium (Gable & Lutz, 2000).

Hypotheses

On the basis of previous studies (Ogden et al., 2010; Ogden et al., 2014), we know that there exist race-ethnic disparities in overweight, obesity, and obesity-related disease among children and adolescents in the United States. The main objective of this study is to evaluate the relative roles of neighborhood race-ethnic segregation and individual-level factors (socio-economic status, household size and structure, family history of obesity) as determinants of these disparities. Our main line of inquiry can be framed by two alternative hypotheses. The null hypothesis simply assumes that race-ethnic disparities in overweight, obesity, and obesity-related disease among children and adolescents are fully explained by individual-level factors, such as socio-economic status, household size and structure, and family history of obesity. The alternative hypothesis is that the observable race-ethnicity disparities in overweight, obesity, and obesity-related chronic disease are not fully explained by the aforementioned socio-demographic characteristics and family history but attributable to persistent racial segregation of residential areas in the United States.

Data and Sample

The individual-level data were obtained from the Panel Study of Income Dynamics’ (PSID) Transition to Adulthood (TA) and Child Development Supplement (CDS) studies. The PSID is an ongoing longitudinal survey, begun in 1968, of a representative sample of U.S. individuals and their families. The initial PSID sample consisted of two independent samples: a cross-sectional, stratified national sample and a national sample of low-income families. Both samples that are combined into the PSID sample are probability samples. Thus, the PSID’s initial households constitute a national probability sample of U.S. households as of 1967. It is worth mentioning that the original PSID sample contains too few Hispanics to draw reliable estimates about the U.S. Hispanic population. Moreover, the initial PSID sample is not representative of individuals who immigrated to the United States after 1968. To remedy these shortfalls, the PSID added 2,043 Hispanic households in 1990. Still, the PSID study undersampled post-1968 immigrants, particularly Asians. Due to a lack of sufficient funding, the Hispanic sample was dropped from the PSID study after 1995. As of now, the share of Hispanic and Asian households in the PSID study is disproportionately lower than their percentage of the U.S. population.

The TA study is part of PSID and a longitudinal follow-on to the CDS. The CDS is an in-home interview survey of caregivers (parents or guardians) about children and adolescents aged 5 to 18 years conducted in 1997, 2002-2003, or 2007. All TA participants had a completed family-level baseline interview in the CDS. This means that each TA sample member had information from the primary caregiver interview about the TA individual when the member was a child. The TA study was initiated in 2005 when the oldest CDS respondents reached 18 to 20 years of age. The TA study has subsequently been conducted in 2007, 2009, and 2011. For this study, we use the most recent TA study conducted in 2011 (TA-11). Individuals were included in the TA-2011 sample if they completed a 2011 core PSID 2011 interview and CDS interview either in 1997, 2002-2003, or 2007. The response rate for the TA survey was 92%.

All of our analyses are conducted using the STATA 10.1 software. To account for unequal probability of selection and survey non-response, our analyses are based on weighted data. Furthermore, cases with missing values one or more of the dependent variables were deleted, resulting in a total of 1,931 cases for the analysis. We also dropped 19 cases (approximately 1% of the sample) belonging to other race-ethnic groups rather than non-Hispanic Whites, African Americans, and Hispanics. Hence, the final sample comprised 1,910 respondents. Auxiliary analyses (not shown for parsimony) confirmed that this deletion did not result in any significant change in the means of the dependent variables. Missing values for all continuous individual-level variables were imputed by the Markov Chain Monte-Carlo technique (for more information on Monte-Carlo imputation, see Rubin, 2004). This imputation technique yielded successive simulations of the distribution of missing values, assuming that the data are missing at random.

With the exception of the outcome variables—adolescent overweight, obesity, and health status—and independent variables monitoring residence and marriage statuses that come from TA-11, the CDS data were used in the present study as they provide the most complete information on the key variables of interest. The average age at the time of interview for CDS was 14.2. The sample consists of 52% females. Approximately 50% of the sample is non-Hispanic White, 42% is African American, and 8% Hispanic.

The PSID records each household’s census tract and metropolitan area of residence. Using this information, we appended to each household’s data record information describing the neighborhood characteristic of the census tract. As in most prior work in this area, we use census tracts as our approximation of neighborhoods. Tract-level data were obtained from the American Community Survey (ACS) summary file.

Measures

The dependent variables in our analyses are adolescent adiposity, proxied by weight status (overweight and obese), and the probability of being diagnosed with obesity-related chronic condition (asthma, diabetes, or hypertension). BMI was calculated as an indicator of adolescent adiposity using self-reported height and weight measures. Following the Centers for Dsease Control and Prevention (CDC) guidelines, overweight is defined as a BMI at or above the 85th percentile and lower than the 95th percentile for children of the same age and sex, and obesity is defined as a BMI at or above the 95th percentile for children of the same age and sex. CDC growth charts were used to define sex-specific cut-points for BMI (Krebs et al., 2007; Kuczmarski et al., 2000).

To avoid the observation of spurious associations between race-ethnicity (both as an individual- and neighborhood-level structural predictor) and obesity, this study controlled for different factors related to both. Hence, the individual-level characteristics used as control variables included parental income, educational attainment, and employment status; household structure categorized as living with two parents (reference), single mother, or other family arrangements; number of siblings; family history of obesity; age, sex (reference: male); residence status (living with parents or not); and marital status (reference: not married). For the exception of the latter two predictors (residence and marital status), all independent variables were obtained from the CDS questionnaire.

The Index of Dissimilarity (ID), perhaps the most commonly used measure of residential segregation in American social science since the 1950s, was used to estimate neighborhood residential segregation. Based always on a comparison of two groups (e.g., Whites vs. Blacks), this measure is very easy to interpret. The ID ranges from 0 to 1, and tells us the percentage of a given ethnic that would have to change its residential location to balance out the distribution of these groups across the geographic space. In this study, we used the ID’s of specific minority groups (e.g., Blacks, Hispanics) versus Whites.

It is worth mentioning that neighborhood effects on weight status are small compared with individual effects in the study sample. An analysis of the intraclass correlations (not shown) reveals that 10.9% and 11.2% of the variance in childhood overweight and obesity, correspondingly, is accounted for by differences in the characteristics of the neighborhoods where PSID participants lived. It also means that, owing to lack of statistical power, only a single predictor variable could be entered into the neighborhood-level models while controlling for demographic and family background variables. At the exploratory stage, we considered a range of neighborhood-level predictors available from the most recent ACS tract-level summary file. Among the factors that we considered were (a) the percentage of adults with less than a high school education, (b) the percentage of households with income below the poverty line, (c) the percentage of single-parent households, and (d) median household income. Still, the exploratory analyses (not shown) indicated that the most significant factor at the school level was the ID. Hence, our choice of a neighborhood-level predictor—ID—is not trivial.

Analytic Strategy

Logistic regression models were used to investigate race-ethnicity disparities in overweight, obesity and obesity-related chronic disease in relation to family socio-demographic, genetic, other contextual, and neighborhood-level (residential segregation) correlates. Several sets of multivariate analyses are reported below. Parallel analyses are estimated for all three dependent variables, represented by the odds of being (a) overweight, (b) obese, and (c) diagnosed with obesity-related chronic condition. Households were used as Level 1 units, and census tracts were used as Level 2 units.

Because the health outcomes listed above were measured at two points of time—when participants were children (at CDS) and when they were adolescents (at TA)—separate regression models were designed for children and adolescent outcomes. Models 1 to 3 are identical for both CDS and TA outcomes. The base Model (Model 1) includes only race-ethnicity dummies. Dummies for African Americans and Hispanic race-ethnicity (Level 1 variables) were entered first so that the subsequent variables predicted any variance not accounted for by these variables. Indicators related to socio-economic status—parental income, educational attainment and employment status—are entered in Model 2. Model 3 adds other individual-level controls: family structure and size, family history of obesity, age, and gender. Model 4 predicting health outcomes measured at CDS includes index of dissimilarity. This Model is identical to Model 5 for TA outcomes. Model 4 predicting health outcomes at the time when participants were adolescents (i.e., at the time of the TA survey) includes residence status (whether a participant lived with his or her parents or not) and marital status.

Descriptive Results

Descriptive statistics for the sample, by race-ethnicity, are provided in Table 1. The table documents significant discrepancies between non-Hispanic White, African American, and Hispanic children and adolescents in all three outcomes of interest—prevalence of overweight, obesity, and obesity-related chronic disease. The sample estimates for the prevalence of overweight and obesity are generally in line with the national averages reported earlier (Ogden et al., 2014). Approximately 11% to 12% of non-Hispanic White children (at CDS) were overweight. In contrast, around 17% of non-Hispanic Black and 18% to 19% of Hispanic children were overweight. It is important to note that prevalence of overweight among non-Hispanic White children did not increase between two points of measurement—the CDS and TA. This was not the case, however, for African American and Hispanic children. Although African American prevalence of obesity was up only slightly at the time of TA, the percentage of overweight among Hispanic children had risen 1%. A somewhat similar dynamic was observed with obesity. Although the period between the CDS and TA showed increases of at least 1% in the prevalence of obesity for both African American and Hispanic children, the percentage of obese among non-Hispanic children (at CDS) and adolescents (at TA) was approximately the same. Furthermore, the disparity of 6% to 9% in the prevalence of obesity between non-Hispanic Whites, on one hand, and African American and Hispanics, on the other, was similar in magnitude to the difference in the prevalence of obesity-related chronic disease between these race-ethnic groups.

OPEN IN VIEWER

Table 1.

Descriptive Statistics of the Sample.

	White (n = 955)	Black (n = 782)	Hispanic (n = 153)	All (N = 1,910)
Dependent variables
Overweight at CDS	11.6%	16.7%	18.1%	14.7%
Overweight at TA	11.4%	17.0%	19.3%	14.8%
Obese at CDS	14.9%	21.7%	22.3%	18.4%
Obese at TA	15.0%	23.6%	24.2%	19.4%
Chronic disease at CDS a	11.4%	17.5%	17.2%	14.4%
Chronic disease at TA a	12.5%	20.2%	19.7%	16.3%
Independent variables
Parental income, thousands of dollars (at CDS)	52.1	40.2	46.1	46.6
Mother’s education, years (at CDC)	14.3	12.3	10.9	13.2
Father’s education, years (at CDC)	13.6	12.3	11.6	12.9
At least one parent employed (at CDS)	89.3%	76.5%	87.3%	84.0%
Two-parent household (at CDS)	62.4%	40.2%	52.6%	52.3%
Single-mother household (at CDC)	19.1%	41.6%	27.9%	28.8%
No. of siblings (at CDS)	1.3	1.4	1.7	1.4
Obese parent or grandparent (at CDS)	20.7%	26.7%	22.4%	23.0%
Age, years (at CDS)	14.2	14.3	14.2	14.2
Gender (male)	47.9%	48.4%	48.1%	48.2%
Resided with one or both parents (at TA)	59.4%	68.0%	60.2%	62.9%
Married or cohabited (at TA)	20.3%	14.4%	20.3%	18.0%
Residential segregation (Level 2)
Dissimilarity Index		0.65	0.41	0.52

Note. CDS = Child Development Supplement; TA = Transition to Adulthood.

a

Includes asthma, diabetes, and hypertension.

Race-ethnic disparities were observed not only in the outcome variables, but also in the independent variables. For example, parental income differed noticeably among the race-ethnic groups. Incomes for African Americans and Hispanics were approximately 23% and 12% lower than non-Hispanic White income. Average level of mother and father education was noticeably higher among non-Hispanic Whites than among African Americans and Hispanics. Similarly, non-Hispanic White children were found to be at advantage over their African American and Hispanic peers with respect to parental employment status, living arrangements (the majority of African American children had not resided in two-parent families), number of siblings (Hispanic children had more siblings than the other two race-ethnic groups), and family history of obesity.

Multivariate Results

Tables 2 to 4 present our multivariate analyses that model race-ethnic disparities in the prevalence of overweight, obesity, and obesity-related disease as a function of individual-level controls and race-ethnic residential segregation. Each table reports the difference in odds ratios of being overweight, obese, and having an obesity-related disease between non-Hispanic Whites, on one hand, and African Americans and Hispanics, on the other. For reasons of parsimony, the coefficients for the control measures are not presented. However, these variables generally conformed to expectations derived from the literature, as negative associations were found between the three dependent variables and parental income, educational attainment, and employment status. Living in a single-mother household and having obese parents or grandparents were also found to be statistically significant predictors of overweight, obesity, and obesity-related disease among children and adolescents alike. Most importantly, we found that the ID is an important risk factor of overweight, obesity, and obesity-related disease. The odds of being diagnosed with obesity and obesity-related disease among children and adolescents, regardless of their race-ethnicity, residing in more segregated neighborhoods were higher than among their peers from less segregated neighborhoods. Complete results are available on request from the author.

OPEN IN VIEWER

Table 2.

Predicted Odds Ratios of Being Overweight (≥85% BMI ≤95%) With and Without Controls.

Model	Regression controls included	African American effect	Hispanic effect
Panel A: Overweight as a child (at CDS)
1	Race-ethnicity dummies, no controls	1.47 [1.32, 1.71]	1.49 [1.22, 1.69]
2	Parental income, education, and employment status	1.24 [1.02, 1.48]	1.31 [1.16, 1.48]
3	Family structure, number of siblings, family history of obesity, age, and gender	1.21 [1.00, 1.44]	1.29 [1.06, 1.45]
4	Dissimilarity Index	1.11 [0.89, 1.25]	1.14 [0.92, 0.39]
Panel B: Overweight as an adolescent/young adult (at TA)
1	Race-ethnicity dummies, no controls	1.49 [1.25, 1.76]	1.54 [1.26, 1.84]
2	Parental income, education, and employment status	1.35 [1.15, 1.58]	1.37 [1.13, 1.63]
3	Family Structure, number of siblings, family history of obesity, age, and gender	1.30 [1.10, 1.53]	1.26 [1.04, 1.48]
4	Residence status and marital status	1.25 [1.08, 1.44]	1.24 [0.95, 1.44]
5	Dissimilarity Index	1.06 [0.87, 1.28]	1.18 [0.99, 1.51]

Note. Non-Hispanic White is the reference category; 95% confidence intervals are in brackets. BMI = body mass index; CDS = Child Development Supplement; TA = Transition to Adulthood.

OPEN IN VIEWER

Table 3.

Predicted Odds Ratios of Being Obese (≥95% of BMI) With and Without Controls.

Model	Regression controls included	African American effect	Hispanic effect
Panel A: Overweight as a child (at CDS)
1	Race-ethnicity dummies, no controls	1.63 [1.35, 1.94]	1.71 [1.35, 1.99]
2	Parental income, education, and employment status	1.53 [1.28, 1.81]	1.56 [1.32, 1.83]
3	Family structure, number of siblings, family history of obesity, age, and gender	1.41 [1.18, 1.66]	1.49 [1.29, 1.71]
4	Dissimilarity Index	1.25 [0.99, 1.49]	1.37 [1.12, 1.56]
Panel B: Overweight as an adolescent/young adult (at TA)
1	Race-ethnicity dummies, no controls	1.73 [1.35, 2.04]	1.86 [1.86, 2.13]
2	Parental income, education, and employment status	1.62 [1.35, 1.93]	1.73 [1.71, 2.06]
3	Family structure, number of siblings, family history of obesity, age, and gender	1.49 [1.22, 1.70]	1.55 [1.49, 1.83]
4	Residence status and marital status	1.45 [1.18, 1.69]	1.45 [1.20, 1.68]
5	Dissimilarity Index	1.28 [1.03, 1.50]	1.34 [1.09, 1.51]

Note. Non-Hispanic White is the reference category; 95% confidence intervals are in brackets. BMI = body mass index; CDS = Child Development Supplement; TA = Transition to Adulthood.

OPEN IN VIEWER

Table 4.

Predicted Odds Ratios of Being Diagnosed With a Chronic Condition (Asthma, Diabetes, or Hypertension) With and Without Controls.

Model	Regression controls included	African American effect	Hispanic effect
Panel A: Diagnosed chronic disease as a child (at CDS)
1	Race-ethnicity dummies, no controls	1.71 [1.35, 1.99]	1.63 [1.35, 1.94]
2	Parental income, education, and employment Status	1.56 [1.32, 1.83]	1.53 [1.28, 1.81]
3	Family structure, number of siblings, family history of obesity, age, and gender	1.49 [1.29, 1.71]	1.41 [1.18, 1.66]
4	Dissimilarity Index	1.28 [1.03, 1.50]	1.37 [1.12, 1.56]
Panel B: Diagnosed with chronic disease as an adolescent/young adult (at TA)
1	Race-ethnicity dummies, no controls	1.96 [1.76, 2.23]	1.73 [1.35, 2.04]
2	Parental income, education, and employment status	1.83 [1.61, 2.16]	1.62 [1.35, 1.93]
3	Family structure, number of siblings, family history of obesity, age, and gender	1.65 [1.39, 1.93]	1.49 [1.22, 1.70]
4	Residence status and marital status	1.45 [1.20, 1.68]	1.45 [1.18, 1.69]
5	Dissimilarity Index	1.25 [0.99, 1.49]	1.34 [1.09, 1.51]

Note. Non-Hispanic White is the reference category; 95% confidence intervals are in brackets. CDS = Child Development Supplement; TA = Transition to Adulthood.

Model 1 of Table 2 (Panel A) shows the odds of being overweight when the study participants were children at the time of CDS survey. The results indicate that, in absence of any controls, both non-Hispanic Blacks and Hispanics are approximately 1.5 times more likely to be overweight than non-Hispanic White children. The addition of variables monitoring family socio-economic status in Model 2 results in a noticeable decline in the odds ratios for both minority groups. The further expansion of the base model through the inclusion of family structure, number of siblings, family history of obesity, and other individual-level controls causes the discrepancy between non-Hispanic Whites and minorities to decrease somewhat (see Model 3). The ID is introduced in Model 4 of Panel A. Its effect eliminates the race-ethnic differences in the prevalence of overweight among children. In real terms, the odds ratios decline to 1.14 for African Americans and 1.11 for Hispanics.

Comparing the results presented in Panels A (study population: children) and B (study population: adolescents) of Table 2, it is hard not to notice a high degree of similarity between the base models of both panels. The estimated race-ethnic disparities in the prevalence of overweight are approximately of the same magnitude among children as well as adolescents. That is, as compared with non-Hispanic White children and adolescents, their African American and Hispanic same-age peers are approximately 1.5 times are more likely to be overweight (if none of the controls are accounted for). Put differently, race-ethnic disparities in overweight persist almost unchanged from childhood to adolescence. Quite similarly to the results of Panel A, the inclusion of variables monitoring family socio-economic status significantly alters the odds ratios for both African Americans and Hispanics. In fact, the largest drop in magnitude of the odds ratios is observed between Models 1 and 2 in both panels of Table 2. This indicates that, as expected, parental income, education, and employment status are important predictors of the prevalence of overweight in children and adolescents and, as such, these predictors exert a mediating effect on the relationship between race-ethnicity and weight status. The next two sets of independent variables entered in Models 3 and 4 of Panel B were also found to be important mediators of the race-ethnicity effect on the prevalence of overweight. Put differently, family structure, number of siblings, marital status, and other predictors entered in Models 3 and 4 explained some of the variance in overweight attributed to race-ethnicity. The odds ratios for African Americans and Hispanics declined steady from approximately 1.35 in Model 2 to 1.25 in Model 4. With the inclusion of ID in Model 5 of Panel B, the odds ratio dropped even further, thus rendering the race-based differences in overweight insignificant.

Turning to Table 3 that documents the race-ethnic discrepancies in obesity, we notice that in the base models of both Panels A and B, (a) Hispanic–White disparity is larger than Black–White disparity in obesity, and (b) the odds of being obese are higher among race-ethnic minorities than the odds of being overweight (if compared with non-Hispanic Whites). The logistic regression results presented in Models 1 of Panels A and B of Table 3 show that, in the absence of any controls, African American children and adolescents are more than 1.6 and 1.7 times, respectively, likely to be obese than non-Hispanic Whites. The corresponding numbers for Hispanic children and adolescents are 1.7 and 1.9 times.

Parallel to the results presented in Table 2 (outcome: the odds of overweight), parental income, education, and employment status, the measures of socio-economic status, have a noticeable mediating effect on the relationship between race-ethnicity and weight status. As evidenced by the drop in the odds ratios for African Americans and Hispanics, family structure, number of siblings, family history of obesity, age, and gender, too, explain some of the association between race-ethnicity and obesity. These mediating effects are similar in magnitude in both Panels A and B of Table 3, confirming that a significant share of the association between race-ethnicity and obesity among children and adolescents, is attributable to family socio-economic status, family structure, and other individual-level risk factors. The introduction of the ID in Model 4 of Panel A alters significantly the odds ratios for both African Americans and Hispanics, but this mediating effect of ID is more pronounced in the case of African Americans than in the case of Hispanics. As compared with Model 3 of Panel A, the odds ratio for African Americans in Model 4 decreases 0.16 points, while that for Hispanics 0.12 points. A somewhat similar picture is observed in Panel B that shows the logistic regression models predicting the odds of adolescent obesity. Hence, it is reasonable to suggest that the portion of the race-based difference in overweight and obesity vis-à-vis non-Hispanic Whites that ID explains is greater for African Americans than for Hispanics.

The next set of multivariate regression analyses shown in Table 4 predicts race-ethnic gaps in the prevalence of obesity-related chronic condition (asthma, diabetes, or hypertension) among children (Panel A) and adolescents (Panel B). In the base models of Panels A and B, the estimated odds ratio for both African Americans and Hispanics are strongly significant and range from 1.63 (Hispanic effect: children) to 1.96 (African American effect: adolescents). In essence, if no controls are included in the regression models, the gap between African American and non-Hispanic White adolescents in the prevalence of obesity-related chronic diseases is estimated to be almost twice as large. The corresponding difference between African American and non-Hispanic White children is smaller but still noteworthy—approximately 1.7 times. Although the Hispanic effect is lesser in magnitude, if compared with the African American one, it is still very significant. Hispanic children and adolescents are 1.6 and 1.7 times, respectively, more likely to be diagnosed with asthma, diabetes, or hypertension than non-Hispanic White children and adolescents. The introduction of socio-economic status measures in Model 2 of both Panels A and B significantly reduces the estimated race-ethnic gaps in chronic disease. Generally, as in previous analyses (see Tables 2 and 3), the effects for parental income, educational attainment, and employment status are consistent across models. Likewise, the addition of other individual-level controls in Model 3 of Panels A and B further reduces the aforementioned race-ethnic disparities. Across models, family socio-economic status and other control measures entered in Models 2 and 3 reduce the race-ethnic gaps in chronic disease by approximately 15%. Residence status (living with parents or separately) and marital status at the time of TA entered in Model 4 of Panel B were also found to reduce the race-ethnic gaps in obesity-related disease. However, the effects for residence and marital statuses were uneven across race-ethnic groups. Adjusting for the aforementioned variables lowered the odds ratio for African Americans more (by 0.2 point) than that for Hispanics (by 0.04 point). The same is true about the effect of ID. The adjustment for the ID decreased the odds ratio for African American children (Panel A) and adolescents (Panel B) to 1.28 and 1.25 (or by about 0.2 point), correspondingly. The parallel values for Hispanics were 1.37 and 1.34 (or by about 0.1 point). Consequently, the effects for residence status, marital status, and ID explain a larger share of the African American–non-Hispanic White disparity in obesity-related chronic disease than the Hispanic–non-Hispanic White disparity.