Explaining Obesity Disparities by Urbanicity, 2006–2016: A Decomposition Analysis

Abstract

Objective.

A large, and potentially growing, disparity in obesity prevalence exists between large central metros and less urban U.S. counties. This study examines its key predictors.

Methods.

Using a rich county-year dataset spanning 2006–2016, we conduct a Gelbach decomposition to examine the relative importance of demographic, socioeconomic, environmental, and behavioral factors in shaping the baseline obesity gap and the growth rate over time between large central metros and other counties.

Results.

Predictors included in our model explain almost the entire obesity gap between large central metros and other counties in the baseline year but can only explain approximately 32% of the growing gap. At baseline, demographic predictors explain more than half the obesity gap, and socioeconomic and behavioral predictors explain the other half. Behavioral and socioeconomic predictors explain more than half the growing gap over time whereas controlling for environmental and demographic predictors decreases the obesity gap by urbanicity over time.

Conclusions.

Results suggest policymakers should prioritize interventions targeting health behaviors of non-large central metros residents to slow the growth of the obesity gap between large central metros and other counties. However, to fundamentally eliminate the obesity gap, in addition to improving health behaviors, policies addressing socioeconomic inequalities are needed.

Introduction

In the United States, the prevalence of obesity has increased from 30.5% in 1999–2000 to 42.4% in 2017–2018 (1). Obesity prevalence varies widely by geography, with less urban counties having a higher prevalence (2,3). The higher obesity prevalence likely overwhelms already limited health care resources available in less urban contexts and exacerbates existing health disparities between individuals in more and less urban areas (4). In particular, the obesity prevalence in large central metros, defined by the National Center for Health Statistics (NCHS) as inner city areas, has been much lower compared to all other types of counties (5,6). In 2012, the prevalence of obesity in large central metro areas was 25.9% while it was 29%−31.3% in less urban areas (6). In addition, obesity prevalence appeared to have grown more rapidly in less urban areas over time during 2001–2016 (1,7). However, previous studies have yet to examine factors that shape this disparity and its potentially growing trend.

Existing evidence has identified various predictors of obesity prevalence. These predictors can be broadly categorized as demographic, socioeconomic (SES), environmental, and behavioral. Among demographic predictors, Black and Latinx residents, older adults, women, and US-born individuals on average have higher obesity risks (8–11). There are a variety of reasons behind these demographic disparities. For example, obesity risks are higher in general among Black and Latinx populations, largely due to structural racism and its downstream determinants, such as poor access to nutritious food, poor built environments conducive for physical activity, and lack of social capital (12–18). In terms of SES predictors, lower levels of SES are associated with increased obesity risks (12–18). Environmental predictors including exposure to greenness and air pollution also play a role in a community’s obesity levels. Finally, a vast literature has identified physical inactivity, unhealthy diets, and risky health behaviors, such as smoking, heavy drinking, and drug use, as potential behavioral predictors of obesity (19–21). Significant differences in all four categories of predictors exist between large central metros and less urban areas (e.g., 29), and many of these predictors have changed substantially over time. For example, in recent decades, less urban counties have had higher and more extreme poverty, greater income inequality, more rapid family structure changes, greater population decline, more hospital closures, and higher growth among people of color (23–25), all of which may contribute to an increasing prevalence of obesity over time (8–11,16).

Examining which category of predictors is more influential in explaining the obesity disparity between large central metros and less urban counties is essential for policymakers to allocate resources efficiently when designing effective interventions to reduce obesity disparities by urbanicity. Reducing obesity disparities by urbanicity is especially important from a social justice and health equity standpoint. Using county-level data spanning the years 2006–16 and a novel decomposition method, this study examines the relative importance of the four categories of predictors, which are selected based on the existing literature on the predictors of obesity, in shaping baseline obesity disparities and the growth rate over time in large central metros and less urban counties.

Methods

We obtained data for a large number of county-level characteristics from various sources during 2006–16 (other years of data are unavailable for 8 predictors, most of which are behavioral predictors). We first visualize obesity prevalence in large central metros and less urban counties over time during 2006–16. We then apply the Gelbach decomposition method (26) to quantify how much of the obesity gap between large central metros and less urban counties as well as its growth over time is due to the inclusion of the demographic, SES, environmental, and behavioral predictors.

Data

Our main dependent variable is county-level obesity prevalence, a measure of the age-adjusted percentage of adults who are 20+ years old and have obesity. Annual data for obesity prevalence during 2006–2016 are from the Centers for Disease Control and Prevention’s (CDC) Diabetes Atlas. We address changing county boundaries over time following Dwyer-Lindgren and colleagues (see eTable 2 in (27)). We measure status of large central metros for each county following the 2006 NCHS urban code scheme (28) (using the 2013 NCHS scheme gives us almost identical results).

We constructed a rich county-year panel dataset. Details on variables and their sources are in Appendix Table S1. If for a given year a predictor is not available, the average of the two closest available years is used (for 2006 and 2016, the closest proximate year is used). Only a very small percentage of counties have missing observations if a certain predictor is available in a given year, and these counties tend to be those with small populations. Missing patterns are shown in Appendix Table S2.

We selected individual predictors based on the existing literature on the predictors of obesity. Our selection has the majority of predictors with only a handful of potential predictors omitted due to data availability, such as crime rates (29), which overlap with many of our included SES predictors. All predictors considered were classified as one of the following: demographic, SES, environmental, and behavioral predictors. Appendix Table S3 reports descriptive statistics for percent with obesity and the four categories of predictors for large central metros and all other counties.

Demographic predictors include county-level age, gender, racial/ethnic, and foreign-born composition. These county-level percentages are potential confounders of the relationship between urbanicity and obesity (for example, counties with higher percentages of Black and Latinx populations disproportionately concentrate in large central metros, and they tend to have higher obesity rates). Therefore, it is important to account for these demographic characteristics in the model.

SES predictors include county-level unemployment, median household income, poverty, rent burden, pupil-to-teacher ratio, percentage with a Bachelor’s degree, percentage of kids receiving free lunch at schools, percentage of single moms, residential segregation (measured by a dissimilarity index following (30)), percentage of the population without health insurance, number of physicians per thousand population, and outpatients per thousand persons. Environmental predictors include Particulate Matter 2.5 score (i.e., PM levels less than 2.5 microns) and high intensity land cover (DHI, an alternative measure for greenspace, captures proportion of a county’s land covered by areas where people live and work, such as apartments, homes, and industrial buildings (31)).

Behavioral predictors include drug prescription rates (opioid prescriptions per 100 residents), percent physically inactive, percent smoking, and percent heavy drinking (based on the average daily consumption of alcohol, measured as more than one drink a day for women and more than two drinks a day for men (32)). In this category, we also included recreational context (number of fitness and recreational centers per thousand persons) and food environment (grocery stores, supercenters, convenience stores, specialized food stores, fast food restaurants, and full-service restaurants per thousand population as well as the percentage of county residents with low access to a grocery store), as the former is closely related to physical exercise and the latter is closely related to diet. We included them in the behavioral instead of the SES category to be consistent with the WHO’s Commission on Social Predictors of Health framework (33), in which health behavior related local contexts and health behaviors are both considered intermediary determinants through which structural determinants (e.g., SES) affect individual health.

Statistical analyses

Two equations clarify our decomposition approach:

$$y_{ct}=\alpha +{\beta }_r*{\mathit{\text{NonLCMetro}}}_c+{\beta }_y*{\mathit{\text{Year}}}_t+{\beta }_{ry}*{\mathit{\text{Year}}}_t*{\mathit{\text{NonLCMetro}}}_c+{\epsilon }_{ct}$$ (1)

$$y_{ct}=\alpha +{\beta }_R*{\mathit{\text{NonLCMetro}}}_c+{\beta }_Y*{\mathit{\text{Year}}}_t+{\beta }_{RY}*{\mathit{\text{Year}}}_t*{\mathit{\text{NonLCMetro}}}_c+{\mathit{\beta }}_{\mathit{D}}\ast {\mathit{D}}_{\mathit{c}\mathit{t}}+{\mathit{\beta }}_{\mathit{S}\mathit{E}\mathit{S}}\ast {\mathit{S}}_{\mathit{c}\mathit{t}}+{\mathit{\beta }}_{\mathit{E}}\ast {\mathit{E}}_{\mathit{c}\mathit{t}}+{\mathit{\beta }}_{\mathit{B}}\ast {\mathit{B}}_{\mathit{c}\mathit{t}}+{\epsilon }_{ct}$$ (2)

where in Equation (1), y_ct indicates the obesity prevalence in county c in year t; β_r is the difference in obesity rates between large central metros and all other counties in the baseline year 2006; β_y represents the change in obesity rates associated with each additional year; β_ry indicates whether the obesity disparity by urbanicity changed over time; and ε_ct are county-level random errors in year t. All models are weighted using county population counts to improve the precision of our estimates. Standard errors are clustered at the county level.

Equation (2) includes all the variables in Equation (1) and adds vectors of demographic (D_ct), SES (S_ct), environmental (E_ct), and behavioral predictors (B_ct). β_R, β_Y, and β_RY in Equation (2) are the coefficients for non-large central metro status at the baseline, a linear year trend, and the differences in obesity trends by urbanicity, respectively, after adjusting for the predictors. β_r − β_R thus indicates the difference in the obesity gaps between large central metros and all other counties in 2006 (baseline) before and after adjusting for the four categories of predictors. β_ry − β_RY indicates the difference in the obesity trends by urbanicity over the study period before and after adjusting for the aforementioned county-level predictors. We considered potential nonlinear relationships between the continuous predictors and the dependent variable (i.e., percent with obesity) by including quadratic terms of continuous predictors when a nonlinear relationship was detected (i.e., when a quadratic term is significant at the 10% level). To avoid issues of multi-collinearity and model overfitting, we carefully examined the correlation matrix of all predictors and relied on BICs for model selection.

Using the Gelbach decomposition method, we partitioned each difference, β_r − β_R and β_ry − β_RY, into proportions that can be attributed to each individual predictor, based on the mathematical derivations in Gelbach (2016) (26). The contributions of the individual predictors can then be aggregated into the four categories: demographic, SES, environmental, and behavioral predictors. This decomposition can inform us which category of predictors is most important for influencing the obesity gap by urbanicity in the baseline and the gap in obesity trends by urbanicity.

Compared to the popular approach that first estimates a model using urbanicity as the predictor without controls, adds each category of controls (in our case, demographic, SES, environmental, and behavioral predictors) sequentially to observe the changes in the urbanicity coefficient, and then attributes the change to the most recently added category, the Gelbach decomposition is not subject to bias caused by the sensitivity of the order in which each category is entered in the model, particularly in our case when the four categories are intercorrelated (26). Compared to the popular Kitagawa-Blinder-Oaxaca (or Oaxaca-Blinder) decomposition, the Gelbach method is based on the omitted variables bias formula, offers a straightforward interpretation of the relative contribution to the overall change, and more easily reports standard errors to quantify the uncertainty (26). In addition, among the four categories of predictors, demographic predictors are generally considered confounders whereas the other three are generally considered mediators in the relationship between urbanicity and obesity prevalence. Compared to the traditional approach of mediation analysis, which relies on the assumption of no confounders, the decomposition approach allows examinations of both confounders and mediators by telling us the proportions of the difference in obesity by urbanicity that is attributable to the differences in the four categories of predictors by urbanicity (34).

Results

Descriptive patterns

Figure 1 graphs the age-adjusted obesity prevalence by urbanicity over 2006–2016. The obesity prevalence in large central metros has been lower compared to less urban counties. In 2006, the gap was 2.36% and it continued to grow over time during 2006–2016. In 2016, this gap increased to 4.46%. In Appendix Figure S1, we further break down the non-large central metro category, by plotting the trends of obesity prevalence for large central metros, large fringe metros, medium metros, small metros, micropolitan counties, and noncore counties. In general, obesity prevalence appeared to be negatively correlated with the level of urbanization, but all types of counties had an increasing obesity prevalence over time. In particular, the obesity prevalence in large central metros has been much lower compared to all other types of counties throughout the period.

Figure 1.

Annual county-level obesity prevalence in large central metros and less urban counties, 2006–2016

Note: We use the 2006 National Center for Health Statistics urban code scheme to assign counties’ urbanicity status.

Decomposition Results

Table 1 presents coefficients of key variables from Equations 1 (Columns A and B) and 2 (Columns C and D). The last column shows the absolute difference between coefficients on urbanicity, the linear time trend, and the interaction term before and after controls were included. Column A indicates that the obesity prevalence in non-large central metros was 2.36 percentage points higher than in large central metros in 2006. Each additional year of increase between 2006–2016 was on average associated with an increase in obesity prevalence of 0.12 percentage points within a large central metro county. The annual increase for non-large central metros is 0.32 (i.e., 0.12+0.2) percentage points. Notably, the positive coefficient of the interaction term shows that the obesity prevalence in non-large central metros increased at a quicker rate over time, relative to large central metros.

Table 1.

Estimates of the obesity gap between large central metros and all other counties, 2006–2016

	Model 1		Model 2
Variable	Coefficient	P-value	Coefficient	P-value	Difference (absolute diff.)	% Total obesity gap explained
	(A)	(B)	(C)	(D)	(A)- (C)	(E)
Non-large central metros (Ref.: large central metros)	2.36	0.00	−0.04	0.88	2.40	100%
	(0.67)		(0.30)
Year Trend	0.12	0.00	0.34	0.00	0.22
	(0.03)		(0.04)
Non-large central metros * Year Trend	0.20	0.00	0.14	0.00	0.07	32%
	(0.03)		(0.03)
Constant	24.41	0.00	−47.46	0.00	71.87
	(0.64)		(16.39)
Controls Included	No		Yes
N (county-years)	32,384

Note: Estimates show coefficients from a Gelbach decomposition with obesity rates as the main outcome; Eq. (1) is shown in model 1 and Eq. (2) is in Model 2 (we use the b1x2 command in STATA). Standard errors are in parentheses and are clustered at the county level. The number in the absolute difference column is not always equal to the difference between column (A) and column (C) because of rounding errors. Same rounding errors apply to column (E). Controls include all socioeconomic, environmental, behavioral, and demographic predictors.

Once county-level demographic, SES, environmental, and behavioral predictors were included, the coefficients on key variables changed markedly. Comparing estimates in Column A and C of Table 1 indicates that the coefficient of non-large central metros drastically decreased from 2.36 to −0.04 and lost its statistical significance. This suggests that predictors included in our model can explain the whole baseline difference in obesity prevalence between large central metros and less urban counties. The coefficient on the time trend, which captures the yearly trend for large central metros—increased and maintained the same direction: Each additional year of increase was associated with a 0.34 percentage point increase in obesity rates. This implies that had the distributions of covariates—either positively or negatively related to obesity—not changed over the study period, obesity would have increased more rapidly in large central metros. Lastly, the coefficient on the interaction term decreased (from 0.20 to 0.14) and remained statistically significant, indicating that obesity rates increased more rapidly in less urban counties compared to large central metros even after controlling for the four categories of predictors. The four categories of predictors explained approximately 32% of the widening obesity gap over time, as shown in Column E.

Table 2 indicates how much each category of predictors contributed to the difference between large central metros and less urban counties in the base year and over time. In 2006, demographic predictors accounted for the largest proportion (52.63%) of the difference. SES predictors accounted for 24.86% of the gap between large central metros and less urban counties, suggesting that less urban counties had lower SES compared to large central metros, which leads to a higher obesity prevalence. Behavioral predictors also played an important role: they accounted for 28.99% of the gap. Environmental predictors explained –6.48% of the difference, consistent with previous literature documenting the relatively better environment in less urban counties than in large central metros. However, this coefficient was not statistically significant at the 10% level, suggesting environmental predictors are not predictive of the obesity gap.

Table 2.

Decomposition of the explained obesity gap between large central metros and less urban counties, 2006–2016

Variable	Non-large central metros (base year: 2006)			Non-large central metros * Year
	Coef.	% Explained gap	% Total gap	Coef.	% Explained gap	%Total gap
Demographic	1.27**	52.63%	52.63%	−0.01**	−19.08%	−6.17%
	(0.41)			(0.01)
Socioeconomic status	0.60*	24.86%	24.86%	0.02 +	26.04%	8.42%
	(0.28)			(0.01)
Environmental	−0.16	−6.48%	−6.48%	−0.03 +	−41.15%	−13.31%
	(0.25)			(0.01)
Behavioral	0.70*	28.99%	28.99%	0.09***	134.19%	43.4%
	(0.30)			(0.02)
Total	2.40***	100.00%	100.00%	0.07**	100.00%	32%
	(0.69)			(0.02)
N (county-years)	32,384

Note: Estimates show the detailed Gelbach decomposition results from Eq. 2 when predictors in Table 1 are grouped into four groups: demographic, socioeconomic, environmental, and behavioral predictors. Standard errors are in parentheses and are clustered at the county level. “% explained gap” refers to the percentage of the contribution of each category of predictors divided by the total contribution of the four categories of predictors in explaining the obesity gap (100% of the total gap in the baseline and 32% of the total increasing gap over time).

⁺p<0.10,

^*p< 0.05,

^**p < 0.01,

^***p < 0.001.

Figure 2 presents each individual predictor’s contribution (in terms of percentage of the total change) to the obesity gap in 2006 between large central metros and less urban counties. Positive coefficients indicate that a given predictor explained a portion of the obesity gap in 2006, conditional on all other predictors. If the coefficient is below zero, the obesity gap would have been larger if large central metros and less urban counties had comparable values in a given predictor. The bar patterns correspond to the four categories of predictors to which each predictor belongs. For example, percent foreign-born accounted for the largest portions (over 70%) of the obesity gap in 2006, followed by percent with a Bachelor’s degree (18%), and percent inactive (18%). These results are consistent with previous literature documenting the higher percentages of immigrants, highly-educated individuals, and individuals who exercise regularly in large central metros compared to less urban counties, which are negatively associated with obesity risks.

Figure 2.

Decomposition results: Contribution to the obesity gap between large central metros and all other counties in 2006

Note: X-axis represents the amount of the obesity gap between large central metros and other counties in the base year 2006 accounted for by each predictor.

Results on the decomposition of the growing obesity gap over time are presented in Table 2 under the column “Non-large central metros * Year”. Notably, behavioral predictors accounted for 43.4% of the total widening obesity gap over time. SES predictors accounted for 8.42% of the total widening obesity gap over time. Demographic and environmental predictors represented −6.17% and −13.31%, respectively, of the total widening obesity gap over time. Adding up the contributions of behavioral and SES predictors, approximately 51.82% of the total widening obesity gap can be accounted for by these two categories of predictors.

Figure 3 shows the contribution of each predictor to the widening obesity gap over time. Percent inactive accounted for the largest proportion (around 54%), followed by percent smoking, the number of convenience stores per 1,000 persons, percent with a Bachelor’s, and various indicators of recreational and food environments. A further investigation of our data shows that percent inactive decreased slightly over time for large central metros, but we did not observe such a pattern for less urban counties. Similarly, the average smoking prevalence slightly decreased more for large central metros over time than for less urban counties. In 2006 and 2007, there was a slightly greater presence of convenience stores in large central metros than in other counties, which was associated with higher obesity risks (35). However, the presence of convenience stores decreased substantially over time in all counties, but the magnitude of the decrease was substantially greater in large central metros. All the evidence suggests that behavioral predictors were improved much more in large central metros than in less urban counties, which contributed to the widening obesity gap over time.

Figure 3.

Decomposition results: Contribution to the obesity gap between large central metros and all other counties, 2006–2016

Note: X-axis represents the amount of the widening obesity gap over time between large central metros and other counties accounted for by each predictor.

Discussion

While obesity rates were increasing at every level of urbanicity along the continuum, we found that the obesity gap between large central metro and less urban counties widened during 2006–16. This gap in obesity rates increased by 2.1 percentage points, from 2.36 percentage points in 2006 to 4.46 percentage points in 2016. Our model’s demographic, SES, and behavioral predictors can explain almost the entire obesity gap between large central metros and other counties in 2006. Among the four categories of predictors, demographic predictors explained more than half of the obesity gap in 2006, and SES and behavioral predictors explained comparable proportions of the obesity gap (24.86% and 28.99%, respectively). Percent foreign-born, percent with a Bachelor’s degree, and percent inactive are the top contributors. However, the four categories of predictors included in the model can only explain approximately 32% of the widening obesity gap over time. Among the four categories of predictors, behavioral predictors explained most of it (43.4% of the total gap), followed by SES predictors (8.42% of the total gap), whereas demographic and environmental predictors made the obesity gap by urbanicity decrease over time. Percent inactive, percent smoking, and the number of convenience stores per 1,000 persons are among the top contributors.

Behavioral predictors appear to play an important role in shaping both the baseline obesity gap and the widening obesity gap over time between large central metros and other counties. It is puzzling that the comprehensive predictors included in our analyses can only explain 32% of the widening obesity gap over time between large central metros and other counties. There are a couple of potential reasons for this finding. First, there are possibly lagged effects of predictors in earlier years. Second, although we have included many county characteristics that are important predictors of obesity based on existing literature, it is possible that these variables are not exhaustive.

SES predictors accounted for approximately 24.86% of the baseline obesity gap between large central metros and other counties. This result highlights the importance of the “fundamental causes” of health (36), which may affect an individual’s access to resources and tend to be associated with various disease outcomes via multiple psychosocial, physiological, and behavioral mechanisms. Consistent with this point of view, we find that in the long-term, as reflected in the baseline disparities, SES predictors have a sizable impact on obesity independently of behavioral factors, because the association between SES predictors and health can be maintained through alternative mechanisms even if the behavioral mechanisms weaken.

Limitations

First, our analyses rely on ecological data, which limit our ability to examine causal effects. In particular, it is possible that the growing obesity gap over time between large central metros and less urban counties is driven by selective migration, if people without obesity disproportionately move to large central metros from less urban counties, and/or people with obesity disproportionately move to less urban counties from large central metros. However, existing evidence shows that people with obesity are less likely to move to a different county in general, but all migrants, regardless of whether they have obesity, like to move to urban counties (37). In addition, Americans tended to move away from both large central metros and rural counties during our study period, to other non-large central metros (38) during the study period. Given the lack of evidence supporting the selective migration concern, our county-level analyses still provide important information to guide public health actions, given that most US local governments are at the county level. Second, due to data availability, we cannot examine data beyond the 2006–16 period. Third, due to space limits, our analysis did not distinguish between different obesity levels, such as overweight, obesity, and Class 3 obesity. Fourth, we used a continuous year variable and thus focused on decomposing the linear trend in the obesity gap. The observed trends in Figure 1 suggest that there may be a nonlinear component. However, decomposing the linear trend allows us to assess how a predictor may contribute to the growing obesity gap. In contrast, nonlinear components are usually not as readily interpretable as the linear term. Finally, due to data availability, a handful of potential predictors were omitted in our model, such as a county’s walkability (39, 40), extent of car-centric transportation (41, 42), and crime rates (29). However, all of these omitted predictors overlap with many of our included predictors to some extent (e.g., DHI, percent inactive, percent in poverty, etc.). In addition, adding proxy measures to a county’s walkability and extent of car-centric transportation in the model, including percent walking to work and percent commuting to walk by car in a county, makes little difference to our results and our model selection results based on BICs suggest we should leave these two measures out. Future studies with available data are needed to further examine the roles of the omitted predictors.

Public Health Implications

The growing obesity gap over time between large central metros and less urban counties is concerning, increasing the already heavy health care burden in less urban counties, especially rural areas. And this trend is concerning even if the gap is partly driven by selective migration. Lee (37) shows that even after accounting for selective migration, rural residence leads to higher odds of obesity. In other words, people who stayed in rural counties, mostly those already with obesity, will continue to be adversely affected by the rural residence. Therefore, regardless of whether the growing trend is driven by selective migration, it is important to have interventions on obesity in rural counties. In addition, even if selective migration is driving part of the trend, when the proportion of people with obesity in less urban counties becomes larger, they may gain weight faster due to peer/network effects (i.e., lack of people without obesity to motivate them to lose weight). Moreover, rural depopulation, especially among healthy individuals, leads to less medical infrastructure in rural areas, but the number of people with health needs stays steady (43, 44). Less medical infrastructure may result in longer distances to hospitals and clinics, and rising incidents of chronic conditions, such as cardiovascular diseases (45, 46).

While existing studies have focused on the rural-urban obesity gap, our findings showed substantial heterogeneity within urban counties, particularly between large central metros and other metro counties. This heterogeneity is particularly salient when examining time trends. Our results also suggest that to narrow the increasing obesity gap over time between large central metros and less urban counties in the short term, policy efforts are needed to improve health behaviors, including making local contexts more amenable to healthy lifestyles, among non-large central metro residents. These efforts should target residents’ exercise, smoking, and dietary behaviors. However, to fundamentally eliminate the obesity gap between large central metros and other counties, in addition to improving health behaviors, policies addressing socioeconomic inequalities are needed. Results in this study can help policymakers prioritize resources under limited budgets.

STUDY IMPORTANCE.

What is already known on this subject?

In the United States, the prevalence of obesity has increased rapidly over the past decades. Obesity prevalence varies widely by geography, with less urban counties having a higher prevalence. In particular, the obesity prevalence in large central metros has been much lower compared to all other types of counties.

What does this study add?

Policymakers should prioritize interventions targeting the health behaviors of residents in non-large central metros to slow down the growth of the obesity gap between large central metros and other counties. However, to fundamentally eliminate the obesity gap, in addition to improving health behaviors, policies addressing socioeconomic inequalities are needed.