Living in High-SES Neighborhoods Is Protective against Obesity among Higher-Income Children but Not Low-Income Children: Results from the Healthy Communities Study

Abstract

Numerous studies have focused on the role of neighborhood socioeconomic status in childhood obesity and physical activity, but few studies have examined the effect of neighborhood socioeconomic changes over time and the interaction between family and neighborhood SES on childhood obesity and physical activity. This study measured neighborhood socioeconomic histories between 2000 and 2010 and examined the associations between neighborhood socioeconomic histories and childhood obesity, as well as physical activity. The moderating role of family poverty status was also examined. Using the Healthy Communities Study (2013–2015), we measured obesity indicators (objectively measured body mass index z-score and waist circumference) and a physical activity indicator (self-reported moderate-to-vigorous physical activity) for a cohort of 4114 children. Multilevel linear regression models were used to examine the associations between neighborhood socioeconomic histories between 2000 and 2009–2013 and body-mass index z-score, waist circumference, and moderate-to-vigorous physical activity. Results showed that higher-income children in consistently high socioeconomic neighborhoods had lower measured BMIz and WC and engaged in more moderate-to-vigorous physical activity than higher-income children in consistently low socioeconomic neighborhoods. Additionally, low-income children in consistently moderate socioeconomic neighborhoods reported a lower level of moderate-to-vigorous physical activity than low-income children in consistently low socioeconomic neighborhoods. The findings indicate that considering both family and neighborhood socioeconomic status may help elucidate the underlying differences in childhood obesity and physical activity levels by socioeconomic status.

Electronic supplementary material

The online version of this article (10.1007/s11524-020-00427-9) contains supplementary material, which is available to authorized users.

Introduction

The role of neighborhood context in childhood obesity has been recognized in research [1–4] and is an area of research that has grown rapidly since 2000. However, there is not a consensus on the relationship between neighborhood socioeconomic status (SES) and physical activity (PA) and more distal outcomes like childhood obesity. Some studies have shown that higher neighborhood SES is associated with a greater level of PA [5, 6], which is an obesity-related behavior [5–7], and a lower risk of childhood obesity independent of the effect of family SES [8–19]. On the other hand, other studies have concluded that there is no significant association between neighborhood SES and childhood obesity [20–25] or PA [21, 26]. These mixed findings suggest that a further exploration of neighborhood SES and its impact on child obesity and PA is needed.

Aside from differences in samples and methods, methodological issues may contribute to the conflicting results. Research examining the impact of neighborhoods on childhood obesity or PA has often relied on a single point-in-time measure of neighborhood SES (such as neighborhood poverty rate in 2000) [8–22, 24, 25, 27, 28]. Yet, a single point-in-time measure ignores potential changes in neighborhood SES over an extended period. For example, the single point-in-time measure of high poverty neighborhoods in the American Community Survey 2005–2009 consisted of neighborhoods with consistently high poverty over decades (39%) and neighborhoods with changing in poverty levels (61%) [29]. In addition, the socioeconomic distribution of residents within neighborhoods has greatly changed, as nearly a fifth of economically disadvantaged neighborhoods in the 50 largest US cities experienced gentrification since 2000 (more than double the rate of the 1990s) [30, 31]. Thus, using a single point-in-time measure fails to account for the dynamic nature of neighborhood SES and results in possible measurement error [29, 32].

Theoretically, neighborhoods with consistently high SES over the past decade may have different social (e.g., safety, social cohesion) and built environmental characteristics (e.g., access to parks, recreation equipment, grocery stores, quality health care) from neighborhoods that recently became high SES neighborhoods as a result of dynamic social, economic, and political circumstances [33, 34]. Neighborhoods with changing SES also face stressors (e.g., rising rents/taxes, decrease in social cohesion) and benefits (e.g., safer streets, cleaner parks) compared to neighborhoods with static SES over time [34]. For example, gentrification increases neighborhood SES and creates a different sense of neighborhood by decreasing affordable housing and forcing displacement, but also improving access and quality of retail and public services, and creating safer streets [30, 31, 35, 36].

Empirical studies have examined the impact of neighborhood SES changes on health and healthy home environments. Do (2009) used data from the Panel Study of Income Dynamics and found that a long-term neighborhood poverty measure is associated with self-rated health and is a stronger predictor of self-rated health than a single point-in-time measure [32]. Walsemann et al. (2017) used data from the Geographic Research and Wellbeing (GROW) study and found mothers living in neighborhoods with decreasing poverty had a lower likelihood of depressive symptoms than those in neighborhoods with consistently low poverty [37]. Another study of the GROW data showed that families in neighborhoods with an increasing median household income had higher odds of healthy home food environments than those in neighborhoods with a decreasing median household income [38]. Similarly, an analysis of the Maternal and Infant Health Assessment showed women in living in neighborhoods with consistently moderate poverty, consistently high poverty, or increasing poverty had higher odds of preterm birth compared to those in neighborhoods with consistently low poverty [29]. This same study found an insignificant association between neighborhood poverty and preterm birth when a single point-in-time measure of neighborhood poverty was used [29]. In support of the theoretical framework and past studies, measuring neighborhood SES histories may better distinguish the effect of living in neighborhoods with static SES on child obesity and PA from that of living in neighborhoods experiencing changing SES.

Another factor that may contribute to conflicting results is that few studies have examined disparate neighborhood effects on obesity or PA among children from low-income families (hereafter, referred to as low-income children) and those from higher income families (hereafter, referred to as higher-income children). According to relative deprivation models [39–44], low-income children in middle or high SES neighborhoods (hereafter, high SES neighborhoods) may experience psychological distress when they compare their family’s SES and their higher-income peers in the same neighborhood. In turn, this may negatively influence health. In addition, even in high SES neighborhoods, health-promoting resources and services could be concentrated where higher-income children reside while health-compromising environments could be located where low-income children reside [45]. Furthermore, due to limited family economic resources, low-income children in high SES neighborhoods may not be able to afford fee-based resources and services provided, publicly and privately, in the neighborhood. Collectively, due to relative deprivation, unbalanced distribution of resources in the neighborhood, and limited family economic resources, living in a high SES neighborhood is not expected to reduce the risk of obesity nor increase children’s engagement in PA for low-income children.

To advance knowledge about neighborhood effects on childhood obesity and PA, the current study measured neighborhood SES histories between 2000 and 2010 and examined the association between neighborhood SES histories and PA, as well as childhood obesity. The moderating effect of family SES on these relationships was also examined.

Methods

Participants

Individual-level data were obtained from the Healthy Communities Study (HCS). The HCS is an observational study of a diverse sample of 130 communities (defined by high school catchment areas) across the USA. The HCS was originally designed to investigate how community programs and policies affect childhood obesity. After stratifying by race, ethnicity, income, region, and a pre-selection score of program and policy intensity, a probability-based sample of 102 high school catchment areas was selected [46, 47]. Twenty-eight high school catchment areas were purposively selected for their childhood obesity prevention efforts [46, 47]. The HCS was designed to oversample high school catchment areas with high proportions of African American or Hispanic populations due to their increased risk for childhood obesity. A total of 5138 children in kindergarten through 8th grade and their parent or guardian participated in the HCS. Data collection occurred from November 2013 through July 2015. The data were collected at home visits by trained field data collection personnel and include anthropometric measurements and socio-demographic information. Additional details about the HCS have been reported elsewhere [47–49].

Our research team extended the HCS by obtaining data on neighborhood socioeconomic context from Census 2000 and the American Community Survey 2009–2013, spatially linking these contextual data to HCS participant data, and examining the associations between neighborhood SES histories and childhood obesity and PA. This study was reviewed by the University of Michigan Institutional Review Board and the Battelle Memorial Institute Institutional Review Board.

Participants were excluded if they moved to the in-home visit address within the past year (n = 274), if they did not have a geocoded address (n = 56), if they had no anthropometric measurement (n = 18) or had a measurement issue on weight and/or height (n = 271), or if they were missing self-reported PA levels (n = 4), family income (n = 295) data, or any other covariates (n = 106). The final sample included 4114 child-parent/guardian dyads. Compared with respondents who remained in the study, those excluded were more likely to be non-Hispanic black, to have less educated parents, and to live in consistently moderate SES neighborhoods.

Measures

Outcome Measures

Outcome measures are two obesity indicators (body mass index z-score (BMIz), waist circumference (WC)) and one PA indicator (moderate-to-vigorous physical activity (MVPA))—a PA indicator is included as a hypothesized obesity-related health behavior [7].

Trained field data collectors measured height, weight, and WC on each child participant during in-home visits between 2013 and 2015. They followed the protocol adapted from the National Health and Nutrition Examination Survey [50] for taking anthropometric measurements. Additional details on anthropometric measures are available elsewhere [51]. We used age- and sex-specific BMIz which is useful for assessing adiposity cross-sectionally [52]. We included WC (in cm) as a supplementary indicator of adiposity [53].

PA was measured using the 7-day Physical Activity Behavior Recall instrument, which asks about participation in 14 activities including physical education, after-school programs, non-school sports, active classes or lessons, and active transport. Parents/guardians (if the child was aged 4–8 years) or children (if the child was aged 9–15 years) reported whether they participated in each activity during the past week, on which days they did the activity, and the average intensity of the activity. The MVPA index was obtained by multiplying the number of moderate-to-vigorous activities reported by frequency of participation in those activities in the past week. Additional details on PA measures are available elsewhere [54].

Independent Variable

The independent variable was neighborhood SES histories. Neighborhood SES was defined as the weighted average of census block group characteristics within 1 km of the participant’s residence. For instance, a participant may have 40% of their 1-km buffer in census block group A, and 60% of their 1-km buffer in census block group B in which case the values for neighborhood SES in census block A would be given a weight of 0.4 while the values for neighborhood SES in census block B would be given a weight of 0.6. Participants are rarely clustered within a neighborhood since each neighborhood is unique to the participant.

First, we measured a single point-in-time measure of neighborhood SES for each period of 2000 and 2009–2013. Neighborhood SES was calculated by summing the z-scores of six variables representing wealth and income (median household income; median value of housing units; and the percentage of households receiving interest, dividend, or net rental income), education (the percentage of adults 25 years of age or older who had completed high school and the percentage of adults 25 years of age or older who had completed college), and occupation (the percentage of employed persons 16 years of age or older in executive, managerial, or professional specialty occupations) [55]. Then, we classified neighborhoods as low-, moderate-, and high-SES by tertiles of SES score for each period. Finally, using SES categories (low, moderate, high) for each period, neighborhoods were categorized into five types of SES histories in a similar way to an a priori method [29]: neighborhoods with consistently low SES; consistently moderate SES; consistently high SES; increasing from low to moderate/high or from moderate to high SES; and decreasing from moderate to low or from high to low/moderate SES.

Covariates

Covariates included child age, child gender, child race/ethnicity, biological parents’ highest education, and neighborhood urbanicity. Neighborhood urbanicity was measured based on the 2010 rural-urban commuting area (RUCA) codes, which classified census tracts based on population density, urbanization, and daily commuting [56]. We categorized the RUCA codes as urban or rural according to the ‘Categorization C’ of WWAMI Rural Health Research Center [57].

Family SES was dichotomized into less than $20,000 and $20,000 or above because the federal poverty threshold in 2013–2015 for a three-person household was $19,530–20,090.

Analyses

Since data were clustered at the level of high school catchment area, multilevel regression modeling was used for the analysis (intraclass correlation = 6.4–8.8% varying by dependent variable). A series of multilevel linear regression models were first conducted for each of the three dependent variables: (model 1) unadjusted models with just one covariate or independent variable and (model 2) a full model with all the covariates and the independent variable. Then, we tested an interaction term between family SES and neighborhood SES histories as a correlate of the dependent variable. Associations between neighborhood SES and child obesity and PA were expected to differ depending on family SES. Where interaction terms were statistically significant at alpha 0.10, we stratified the sample by family SES (< $20,000 or ≥ $20,000) and conducted multilevel linear regression to examine differences in the association between neighborhood SES histories and the dependent variable among low-income children versus higher-income children. In addition, we conducted multilevel linear regression to examine associations between a single point-in-time measure of neighborhood SES from the ACS (2009–2013) and each dependent variable, after adjusting for covariates. Finally, we conducted a final model only among children living in the neighborhood for 3 years or more (n = 2955) as a sensitivity analysis. Analyses were performed using Stata Version 15.

Results

Sample characteristics and descriptive statistics of the childhood obesity outcomes and PA outcome are shown in Table 1. Twenty-nine percent of the sample lived in neighborhoods with consistently low SES, and another 29% lived in neighborhoods with consistently high SES. The sample included 45% who self-identified as Hispanic, 31% who self-identified as non-Hispanic white, 16% who self-identified as non-Hispanic black, and 8% who self-identified as other or multiple racial/ethnic group.

Table 1.

Sample characteristics and descriptive statistics of obesity and physical activity outcomes by sample characteristics, Healthy Communities Study, USA, 2013–2015, N = 4114

	Total sample		Higher-income children (n = 3027)			Low-income children (n = 1087)
	N	%	BMIz	WC	MVPA	BMIz	WC	MVPA
Total sample	4114	100	0.63	69.1	13.5	0.87	71.4	13.3
Neighborhood SES histories
Consistently low	1211	29.4	0.88	72.6	13.3	0.92	72.1	14.3
Consistently moderate	1007	24.5	0.81	70.5	12.6	0.81	70.2	11.8
Consistently high	1202	29.2	0.32	65.5	14.2	0.88	72.8	14.4
Increasing	342	8.3	0.62	69.8	12.6	0.74	70.3	12.7
Decreasing	352	8.6	0.79	70.8	14.3	0.95	71.8	12.6
Child age
4–6 years	747	18.2	0.62	58.1	13.0	0.78	59.0	11.8
7–9 years	1435	34.9	0.61	64.6	14.1	0.82	66.6	12.7
10–12 years	1321	32.1	0.63	74.7	13.7	1.00	78.1	14.2
13–15 years	611	14.9	0.70	81.2	12.5	0.79	83.1	14.5
Child gender
Boy	2047	49.8	0.67	68.8	13.8	0.86	70.8	13.7
Girl	2067	50.2	0.58	69.4	13.2	0.87	71.8	12.9
Child race/ethnicity
Black, non-Hispanic	673	16.4	0.88	70.6	12.1	0.76	68.7	12.9
Hispanic	1863	45.3	0.91	72.4	13.4	0.96	73.1	13.0
White, non-Hispanic	1261	30.7	0.31	65.9	14.0	0.68	69.6	14.7
Other/multiple	317	7.7	0.32	65.8	14.5	0.68	68.5	14.1
Maximum parent education
Less than high school	915	22.2	0.93	73.4	13.3	0.98	72.9	12.7
High school graduate	789	19.2	0.74	71.2	13.3	0.76	69.5	13.4
Some college	1026	24.9	0.84	71.1	13.2	0.80	70.7	14.3
College graduate or above	1384	33.6	0.35	65.6	13.9	0.88	72.5	12.6
Urbanicity
Urban	3681	89.5	0.62	69.0	13.5	0.85	71.5	13.2
Rural	433	10.5	0.75	70.1	13.9	0.95	70.5	13.7

Among higher-income children, BMIz and WC were lower in neighborhoods with consistently high SES (0.32 and 65.5, respectively) than in other types of neighborhoods (0.62–0.88 and 69.8–72.6, respectively). The mean number of times the child engaged in MVPA during the past week was 13.5 among higher-income children and 13.3 among low-income children. Higher-income children in consistently high or decreasing SES neighborhoods reported more frequent MVPA (14.2–14.3) than those in consistently low SES neighborhoods (13.3), consistently moderate SES neighborhoods (12.6), and increasing SES neighborhoods (12.6). Among low-income children, the frequency of MVPA was lower in neighborhoods with consistently moderate SES (11.8) than those living in all other types of neighborhoods (12.6–14.3).

We found the interaction term between family SES and neighborhood SES histories was significantly associated with three dependent variables (results not shown). Accordingly, we stratified the sample by family SES and conducted hierarchical linear regression models. Tables 2, 3, and 4 present the results of hierarchical linear regression examining associations between neighborhood SES histories and BMIz, WC, or MVPA. As shown in the full model in Table 2, among higher-income children, those living in consistently high SES neighborhoods had lower BMIz than those living in neighborhoods with consistently low SES (b = − 0.19, p < .01). The fit of the model was improved from the unconditional model (− 2LL = 9554; AIC = 9559) to the final full model (− 2LL = 9469; AIC = 9473). Table 3 shows that higher-income children in neighborhoods with consistently high SES had lower WC than those in neighborhoods with consistently low SES (b = − 2.92, p < .001). The fit of the model was improved from the unconditional model (− 2LL = 24,615; AIC = 24,619) to the final model (− 2LL = 23,477; AIC = 24,481). Among low-income children, neighborhood SES histories were not significantly associated with BMIz and WC.

Table 2.

Associations between neighborhood SES histories and BMIz, Healthy Communities Study, USA, 2013–2015, N = 4114

	Higher-income children (n = 3027)								Low-income children (n = 1087)
	Unadjusted bivariate models				Full model				Unadjusted bivariate models				Full model
	b	SE	95% CI		b	SE	95% CI		b	SE	95% CI		b	SE	95% CI
Neighborhood SES histories
Consistently low (reference)
Consistently moderate	− 0.06	0.07	− 0.20	, 0.09	− 0.01	0.07	− 0.15	, 0.13	− 0.11	0.09	− 0.28	, 0.06	− 0.06	0.09	− 0.23	, 0.12
Consistently high	− 0.44***	0.07	− 0.58	, − 0.31	− 0.19**	0.07	− 0.33	, − 0.04	− 0.04	0.14	− 0.32	, 0.23	0.02	0.15	− 0.27	, 0.31
Increasing	− 0.19*	0.09	− 0.37	, − 0.01	− 0.06	0.09	− 0.24	, 0.12	− 0.18	0.14	− 0.44	, 0.09	− 0.13	0.14	− 0.40	, 0.13
Decreasing	− 0.06	0.09	− 0.24	, 0.12	0.02	0.09	− 0.15	, 0.19	0.02	0.15	− 0.27	, 0.32	0.06	0.15	− 0.23	, 0.35
Child age
4–6 years (reference)
7–9 years	− 0.02	0.06	− 0.13	, 0.10	− 0.03	0.06	− 0.15	, 0.09	0.04	0.10	− 0.17	, 0.24	0.04	0.10	− 0.17	, 0.24
10–12 years	− 0.02	0.06	− 0.14	, 0.11	− 0.05	0.06	− 0.17	, 0.07	0.21*	0.11	0.00	, 0.42	0.19	0.11	− 0.01	, 0.40
13–15 years	0.06	0.08	− 0.09	, 0.21	0.04	0.07	− 0.10	, 0.18	− 0.01	0.13	− 0.24	, 0.26	− 0.01	0.13	− 0.26	, 0.23
Child gender
Boy (reference)
Girl	− 0.08*	0.04	− 0.17	, 0.00	− 0.10*	0.04	− 0.19	, − 0.02	0.01	0.07	− 0.13	, 0.15	0.02	0.07	− 0.12	, 0.17
Child race/ethnicity
Black, non-Hispanic	0.47***	0.07	0.33	, 0.61	0.38***	0.07	0.24	, 0.53	0.08	0.13	− 0.17	, 0.33	0.09	0.13	− 0.16	, 0.35
Hispanic	0.54***	0.05	0.43	, 0.64	0.43***	0.06	0.31	, 0.55	0.29**	0.12	0.08	, 0.51	0.28*	0.12	0.04	, 0.52
White, non-Hispanic (reference)
Other/multiple	0.06	0.08	− 0.10	, 0.21	0.06	0.08	− 0.10	, 0.21	0.00	0.20	− 0.38	, 0.39	0.04	0.20	− 0.35	, 0.42
Maximum parent education
Less than high school	0.43***	0.07	0.30	, 0.56	0.16*	0.08	0.01	, 0.31	0.09	0.14	− 0.19	, 0.38	− 0.01	0.15	− 0.31	, 0.29
High school graduate	0.27***	0.07	0.14	, 0.40	0.05	0.07	− 0.09	, 0.19	− 0.12	0.15	− 0.41	, 0.17	− 0.15	0.15	− 0.45	, 0.15
Some college	0.39***	0.06	0.28	, 0.49	0.24***	0.06	0.12	, 0.35	− 0.08	0.15	− 0.38	, 0.21	− 0.08	0.16	− 0.39	, 0.22
College graduate or above (reference)
Urbanicity
Urban	− 0.10	0.12	− 0.33	, 0.13	− 0.10	0.09	− 0.27	, 0.07	− 0.11	0.12	− 0.34	, 0.12	− 0.13	0.12	− 0.37	, 0.10
Rural (reference)
Model fit
− 2LL					9469.4								3486.1
AIC					9473.4								3490.1
BIC					9479.1								3495.7

^*p < .05, ^**p < .01, ^***p < .001

Table 3.

Associations between neighborhood SES histories and WC, Healthy Communities Study, USA, 2013–2015, N = 4114

	Higher-income children (n = 3027)								Low-income children (n = 1087)
	Unadjusted bivariate models				Full model				Unadjusted bivariate models				Full model
	b	SE	95% CI		b	SE	95% CI		b	SE	95% CI		b	SE	95% CI
Neighborhood SES histories
Consistently low (reference)
Consistently moderate	− 1.38	0.93	− 3.21	, 0.45	− 1.01	0.75	− 2.45	, 0.45	− 1.73	1.23	− 4.14	, 0.68	− 1.01	1.02	− 3.00	, 0.99
Consistently high	− 5.10***	0.89	− 6.85	, − 3.36	− 2.92***	0.77	− 4.43	, − 1.42	0.89	1.93	− 2.89	, 4.67	− 0.16	1.72	− 3.52	, 3.21
Increasing	− 1.82	1.14	− 4.06	, − 0.42	− 1.41	0.94	− 3.25	, 0.43	− 1.55	1.87	− 5.22	, 2.11	− 0.83	1.57	− 3.91	, 2.24
Decreasing	− 0.77	1.13	− 2.98	, 1.44	− 0.48	0.91	− 2.27	, 1.31	0.31	2.05	− 4.34	, 3.72	− 0.26	1.72	− 3.64	, 3.11
Child age
4–6 years (reference)
7–9 years	6.41***	0.62	5.19	, 7.63	6.35***	0.62	5.14	, 7.55	7.57***	1.22	5.18	, 9.95	7.59***	1.21	5.22	, 9.96
10–12 years	16.33***	0.64	15.08	, 17.57	16.06***	0.63	14.84	, 17.29	19.03***	1.24	16.60	, 21.46	18.95***	1.23	16.54	, 21.36
13–15 years	22.64***	0.77	21.14	, 24.14	22.61***	0.75	21.14	, 24.09	24.07***	1.47	21.18	, 26.95	23.87***	1.46	21.01	, 26.73
Child gender
Boy (reference)
Girl	0.46	0.51	− 0.54	, 1.46	0.36	0.43	− 0.47	, 1.20	1.03	0.98	− 0.89	, 2.95	0.93	0.84	− 0.71	, 2.56
Child race/ethnicity
Black, non-Hispanic	3.81***	0.90	2.06	, 5.57	1.44	0.77	− 0.06	, 2.94	− 0.79	1.72	− 4.16	, 2.59	− 1.63	1.49	− 4.55	, 1.29
Hispanic	5.15***	0.69	3.80	, 6.50	3.42***	0.64	2.17	, 4.66	3.45*	1.56	0.39	, 6.52	2.66	1.41	− 0.10	, 5.42
White, non-Hispanic (reference)
Other/multiple	0.23	0.97	− 1.67	, 2.13	− 0.15	0.81	− 1.74	, 1.44	− 1.02	2.67	− 6.27	, 4.22	− 0.27	2.30	− 4.77	, 4.23
Maximum parent education
Less than high school	5.83***	0.82	4.22	, 7.44	1.84*	0.78	0.31	, 3.36	0.21	1.97	− 3.64	, 4.06	− 1.35	1.79	− 4.85	, 2.15
High school graduate	4.02***	0.80	2.46	, 5.58	0.82	0.72	− 0.60	, 2.23	− 3.08	2.00	− 7.01	, 0.84	− 2.48	1.78	− 5.97	, 1.00
Some college	4.07***	0.67	2.76	, 5.38	2.19***	0.59	1.03	, 3.35	− 1.81	2.05	− 5.84	, 2.22	− 1.13	1.80	− 4.66	, 2.40
College graduate or above (reference)
Urbanicity
Urban	− 1.12	1.34	− 3.75	, 1.52	0.93	0.91	− 0.85	, 2.72	0.60	1.65	− 2.64	, 3.85	− 0.31	1.40	− 3.04	, 2.43
Rural (reference)
Model fit
− 2LL					23,476.9								8737.5
AIC					23,480.9								8739.5
BIC					23,486.6								8742.4

^*p < .05, ^**p < .01, ^***p < .001

Table 4.

Associations between neighborhood SES histories and MVPA, Healthy Communities Study, USA, 2013–2015, N = 4114

	Higher-income children (n = 3027)								Low-income children (n = 1087)
	Unadjusted bivariate models				Full model				Unadjusted bivariate models				Full model
	b	SE	95% CI		b	SE	95% CI		b	SE	95% CI		b	SE	95% CI
Neighborhood SES histories
Consistently low (reference)
Consistently moderate	0.82	0.54	− 0.25	, 1.88	0.97	0.56	− 0.12	, 2.06	− 2.57**	0.78	− 4.09	, − 1.04	− 2.71**	0.81	− 4.30	, − 1.12
Consistently high	1.83***	0.52	0.81	, 2.86	1.79**	0.57	0.68	, 2.91	0.10	1.13	− 2.32	, 2.11	− 0.48	1.20	− 2.83	, 1.88
Increasing	0.41	0.65	− 0.87	, 1.68	0.54	0.66	− 0.76	, 1.83	− 1.16	1.10	− 3.31	, 0.99	− 1.37	1.11	− 3.55	, 0.82
Decreasing	1.74**	0.64	0.48	, 3.00	1.80**	0.65	0.53	, 3.07	− 1.90	1.19	− 4.23	, 0.44	− 2.09	1.20	− 4.44	, 0.26
Child age
4–6 years (reference)
7–9 years	1.22**	0.41	0.43	, 2.02	1.20**	0.41	0.41	, 2.00	1.18	0.78	− 0.35	, 2.72	1.32	0.78	− 0.21	, 2.85
10–12 years	0.62	0.42	− 0.19	, 1.44	0.67	0.42	− 0.15	, 1.49	2.33**	0.80	0.76	, 3.91	2.48**	0.80	0.90	, 4.05
13–15 years	− 0.62	0.50	− 1.61	, 0.36	− 0.64	0.50	− 1.63	, 0.34	2.45*	0.96	0.58	, 4.33	2.59**	0.96	0.71	, 4.46
Child gender
Boy (reference)
Girl	− 0.62*	0.28	− 1.18	, − 0.07	− 0.58*	0.28	− 1.13	, − 0.02	− 0.90	0.54	− 1.96	, 0.17	− 1.01	0.54	− 2.07	, 0.06
Child race/ethnicity
Black, non-Hispanic	− 0.95	0.51	− 1.89	, 0.17	− 0.57	0.53	− 1.60	, 0.47	− 1.24	1.01	− 3.22	, 0.75	− 1.58	1.02	− 3.58	, 0.41
Hispanic	− 0.82*	0.39	− 1.59	, − 0.04	− 0.52	0.44	− 1.39	, 0.34	− 2.08*	0.93	− 3.90	, − 0.27	− 1.83	0.96	− 3.71	, 0.05
White, non-Hispanic (reference)
Other/multiple	0.34	0.54	− 0.72	, 1.41	0.33	0.54	− 0.73	, 1.39	− 0.92	1.50	− 3.87	, 2.03	− 0.57	1.50	− 3.51	, 2.37
Maximum parent education
Less than high school	− 0.77	0.47	− 1.69	, 0.14	− 0.08	0.52	− 1.10	, 0.94	− 0.35	1.11	− 2.53	, 1.83	0.12	1.17	− 2.18	, 2.42
High school graduate	− 0.22	0.45	− 1.10	, 0.66	0.38	0.48	− 0.56	, 1.33	0.68	1.12	− 1.53	, 2.88	1.24	1.16	− 1.04	, 3.51
Some college
College graduate or above (reference)	− 0.57	0.38	− 1.31	, 0.16	− 0.15	0.39	− 0.93	, 0.62	1.29	1.15	− 0.97	, 3.55	1.57	1.18	− 0.74	, 3.88
Urbanicity
Urban	− 0.59	0.75	− 2.07	, 0.88	− 1.01	0.78	− 2.53	, 0.51	− 0.84	1.07	− 2.94	, 1.26	0.38	1.10	− 1.79	, 2.54
Rural (reference)
Model fit
− 2LL					21,003.5								7809.3
AIC					21,007.5								7813.3
BIC					21,013.2								7819.0

^*p < .05, ^**p < .01, ^***p < .001

Table 4 shows that higher-income children in neighborhoods with consistently high SES or decreasing SES engaged in MVPA more frequently compared to those in neighborhoods with consistently low SES (b = 1.79 and 1.80, respectively, p < .01). Among low-income children, those in neighborhoods with consistently moderate SES participated in MVPA less frequently than those in neighborhoods with consistently low SES (b = − 2.71, p < .01). The model fit was improved from the unconditional model (higher-income children: − 2LL = 21,057; AIC = 21,061; low-income children: − 2LL = 7869; AIC = 7873) to the final full model (higher-income children: − 2LL = 21,004; AIC = 21,008; low-income children: − 2LL = 7809; AIC = 7813).

Table 5 presents associations between a single point-in-time measure of neighborhood SES and each dependent variable. Higher-income children living in high SES neighborhoods had lower BMIz than those in low SES neighborhoods (b = − 0.22, p < .01). Higher-income children living in moderate or high SES neighborhoods had lower WC than those in low SES neighborhoods (b = − 1.55, p < 0.05 and b = − 2.82, p < .001, respectively). Among higher-income children, neighborhood SES was not significantly associated with MVPA. Among low-income children, neighborhood SES was not significantly associated BMIz and WC. Low-income children in moderate SES neighborhoods engaged in MVPA less frequently than those in low SES neighborhoods (b = − 2.30, p < 0.01). Finally, a sensitivity analysis was conducted only with children living in the neighborhood for 3 years or more. Similar results were found (see Appendix Table 1).

Table 5.

Associations between a single point-in-time measure of neighborhood SES and BMz, WC, or MVPA among children living in the neighborhood, Healthy Communities Study, U.S., 2013-2015, N=4,114

	BMIz			WC			MVPA
	b	SE	95% CI	b	SE	95% CI	b	SE	95% CI
Higher-Income children
Neighborhood SES in 2009-2013
Low SES (reference)
Moderate SES	−0.09	0.07	−0.23, 0.05	−1.55*	0.78	−3.08, −0.02	0.52	0.57	−0.59, 1.63
High SES	−0.22**	0.08	−0.37, −0.07	−2.82***	0.83	−4.44, −1.20	0.89	0.60	−0.29, 2.07
Low-Income children
Neighborhood SES in 2009-2013
Low SES (reference)
Moderate SES	−0.09	0.10	−0.11, 0.29	0.39	1.21	−1.99, 2.77	− 2.30**	0.86	−3.97, −0.62
High SES	−0.06	0.15	−0.35, 0.22	−0.92	1.74	−4.34, 2.51	−0.57	1.19	−2.89, 1.76

Note. Covariates in Tables 2, 3, and 4 were controlled

^*p < .05, ^**p < .01, ^***p < .001

Discussion

This study examined the associations between neighborhood SES histories and childhood obesity and PA outcomes with consideration of family SES. Previous literature has shown inconsistent findings regarding the associations between a single point-in-time measure of neighborhood SES and childhood obesity and PA (see review paper [58]). Our findings extend previous knowledge by measuring neighborhood SES histories and showing the disparate associations of neighborhood SES histories with obesity and PA between higher-income and low-income children. The inconsistent findings in past literature may be due to the lack of consideration of neighborhood SES histories, the moderating role of family SES in such associations, or both.

Higher-Income Children: Associations between Neighborhood SES and Obesity and MVPA

Higher-income children living in consistently high SES neighborhoods had the lowest BMIz and WC values compared to all other types of neighborhood SES histories. Multivariate results also showed living in consistently high SES neighborhoods was associated with a lower level of obesity indicators than living in consistently low SES neighborhoods among higher-income children with adjustment for sociodemographic characteristics and neighborhood urbanicity. Similar results were obtained for MVPA. Among higher-income children, living in consistently high or decreasing SES neighborhoods was associated with more frequent MVPA than living in consistently low SES neighborhoods, accounting for sociodemographic characteristics and neighborhood urbanicity. Because more than half of decreasing SES neighborhoods had a high level of SES in 2000, decreasing SES neighborhoods tend to be advantaged. The results corroborate empirical evidence supporting that living in socioeconomically advantaged neighborhoods is associated with decreased obesity risk [5, 6] and greater engagement in PA [8–19].

Using a measure of neighborhood SES histories rather than a single point-in-time measure of neighborhood SES, we captured a clearer pattern of associations between neighborhood SES and obesity and MVPA among higher-income children. For example, the high SES category of a single point-in-time measure (based on ACS 2009–2013) consists of consistently high SES neighborhoods and increasing SES neighborhoods (shown in Table 6). Therefore, a single point-in-time measure would not allow us to see the nuanced relationship between neighborhood SES and obesity: Higher-income children living in consistently high SES neighborhoods have lower BMIz and WC values than those living in consistently low neighborhood SES neighborhoods, while children living in increasing SES neighborhoods do not have significantly different BMIz and WC values compared to those living in consistently low SES neighborhoods. As another example, significant associations between neighborhood SES and MVPA among higher-income children were found only when using a measure of neighborhood SES histories. On the other hand, if we had used only a single point-in-time measure of neighborhood SES, we would have concluded that neighborhood SES was not associated with MVPA among higher-income children. Notably, results of models using a measure of neighborhood SES histories may reduce statistical power. Neighborhoods with moderate SES in 2009–2013 (n = 899) could be categorized as consistently moderate SES neighborhoods (n = 657), increasing SES neighborhoods (n = 102), or decreasing SES neighborhoods (n = 140) in a measure of neighborhood SES histories. When using a measure of neighborhood SES histories, none of these categories were significantly associated with WC, which suggests dividing the category of moderate SES into three categories reduces statistical power.

Table 6.

Distribution of neighborhoods categorized by a single point-in-time measure of neighborhood SES and a measure of neighborhood SES histories, Healthy Communities Study, USA, N = 4114

	Consistently low SES	Consistently moderate SES	Consistently high SES	Increasing SES	Decreasing SES	Total
Higher-income children
Point-in-time SES (2009–2013)
Low SES	731 (84.2%)	–	–	–	137 (15.8%)	868 (100.0%)
Moderate SES	–	657 (73.1%)	–	102 (11.4%)	140 (15.6%)	899 (100.0%)
High SES	–	–	1115 (88.5%)	145 (11.5%)	–	1260 (100.0%)
Low-income children
Point-in-time SES (2009–2013)
Low SES	480 (90.2%)	–	–	–	52 (9.8%)	532 (100.0%)
Moderate SES	–	350 (83.5%)	–	46 (11.0%)	23 (5.5%)	419 (100.0%)
High SES	–	–	87 (64.0%)	49 (36.0%)	–	136 (100.0%)

The specific mechanisms through which neighborhood SES histories are associated with obesity and PA among higher-income children is not clear from these data. However, based on theory and past research [59–67], it is likely that built and social environments serve as potential mediators in the associations between neighborhood SES and child obesity or PA. How such mechanisms may play out is an important area for longitudinal research.

Low-Income Children: Associations between Neighborhood SES and Obesity and MVPA

Among low-income children, neighborhood SES histories were not associated with obesity indicators. Neighborhood SES histories were associated with MVPA in a different way from higher-income children. Accounting for sociodemographic characteristics and neighborhood urbanicity, low-income children in consistently moderate SES neighborhoods reported a lower level of MVPA than those in consistently low SES neighborhoods. On the other hand, living in consistently high SES neighborhoods (vs. consistently low SES neighborhoods) was not associated with MVPA. This may be due to power issues related to the low number of low-income children in consistently high SES neighborhoods (n = 87). Our findings are consistent with a previous study findings showing low-income children in higher-poverty and higher-inequality neighborhoods were more likely to report insufficient PA than those in low-poverty and higher-inequality neighborhoods [45]. The results support relative deprivation models [39–45] in that low-income children living in moderate to high SES neighborhoods tend to have more or worse adverse health outcomes compared to those in low SES neighborhoods due to relative deprivation, unbalanced distribution of resources in the neighborhood, and limited family economic resources.

It is important to consider why neighborhood SES histories may not be associated with obesity indicators but associated with MVPA among low-income children. Although MVPA is known to be important in preventing childhood obesity, research indicates that other obesity-related behaviors—such as energy and nutrient intake and sleeping duration—also influence child obesity status [68, 69]. Differential impacts of neighborhood SES by types of obesity-related behaviors may explain the non-significant associations of neighborhood SES with BMIz and WC. Future research is needed to investigate the disparate effects of neighborhood SES on obesity-related behaviors such as dietary habits, PA, sedentary behavior, and active commuting to school.

Using a measure of neighborhood SES histories, we captured that living in neighborhoods with consistently moderate SES, but not neighborhoods with increasing or decreasing SES, is detrimental for promoting MVPA among low-income children. Therefore, a single point-in-time measure would not allow us to see that low-income children living in increasing or decreasing SES neighborhoods do not have significantly different MVPA values compared to those living in consistently low SES neighborhoods.

Limitations

This study was subject to a number of limitations. Although this study considered the length of living in the neighborhood, reverse causation cannot be ruled out in the cross-sectional design used in this study. Namely, those who are physically inactive tend to move to low SES neighborhoods that lack health-promoting built and social environments. MVPA was based on the 7-day Physical Activity Behavior Recall instrument for which the psychometric properties are not established [54]. Also, despite being a national study, the HCS is not a nationally representative sample and its racial/ethnic distribution is different from national racial/ethnic distribution. Therefore, our results cannot be generalized to all other children in the USA. In addition, given the sample size differs depending on neighborhood SES types and family SES, there is a possibility the statistical analyses for low-income children were underpowered. Finally, there are some differences between the design and methodology of the ACS and Census 2000 [70]. While Census 2000 references estimates to a specific date, as in April 1, 2000, ACS period estimates are based on nearly daily collection of information over the entire calendar year. Also, while Census 2000 asked for a respondent’s income during 1999, the ACS asked for a respondent’s income over the past 12 months.

Despite the limitations, this study had several strengths. Neighborhood SES was defined as weighted average of census block group characteristics within 1 km of each residence, which might correspond with people’s perceived neighborhoods [71–73]. Additionally, using a buffer rather than selecting the participant’s block group minimizes any border effects [74]—using block groups may be adequate for residents living at or near the center of the block group, but not for those living near the borders between two or more block groups [75]. Also, using 1-km buffers is congruent with literature showing perceived conditions of neighborhood are most highly correlated within small neighborhoods, such as 1-km buffers [73, 76]. In addition, literature has documented limitations of using a single point-in-time measure, which combines neighborhoods with heterogeneous socioeconomic histories and fails to account for the dynamic social, economic, and political circumstances in the neighborhoods [29, 32, 37]. This study measured neighborhood SES histories between 2000 and 2009–2013 by distinguishing neighborhoods which experienced either increasing or decreasing SES for the past decade from those with consistent levels of SES. We also considered the interaction effects between family- and neighborhood-SES on childhood obesity and PA, which has been theoretically discussed but not often tested in empirical research (see for exception [45]). Finally, WC was employed as a supplementary obesity indicator, which provides insight into the distribution of adiposity better than BMIz alone [51].

Conclusions

Our findings emphasize the importance of considering neighborhood SES histories and the interactions between family- and neighborhood-SES in work examining childhood obesity and PA. In summary, living in consistently high SES neighborhoods benefits obesity prevention for higher-income children, but not for low-income children. In addition, living in consistently moderate SES neighborhoods (vs. consistently low SES neighborhoods) hinders low-income children from engaging in MVPA, which is an obesity-related health behavior. Future studies should consider utilizing neighborhood SES histories in order to better elucidate the unique and complex influences of current and historical neighborhood SES on health behaviors.

The findings have important implications. Policy makers and stakeholders need to consider additional programs and policies for children in socioeconomically disadvantaged neighborhoods (i.e., consistently low, consistently moderate, increasing, or decreasing SES neighborhoods) to combat the negative effects that neighborhood SES has on obesity in these subpopulations. In addition, in order to promote PA, community policies and programs need to consider the differential experiences of low-income children in consistently moderate neighborhoods and tailor health-promoting programs and environments to address these differences. For example, it is critical to ensure neighborhood resources are at reasonable costs for low-income families.

Compliance with Ethical Standards

Ethics Approval

This study was reviewed by the University of Michigan Institutional Review Board and the Battelle Memorial Institute Institutional Review Board.