The interactive role of SNAP participation and residential neighborhood in childhood obesity

ABSTRACT

Nationally representative studies of childhood obesity have examined the roles played by neighborhood conditions and by SNAP use, but not the effects of these factors together or in interaction. We used restricted, geocoded data from the 1986–2012 Child and Young Adult sample of the National Longitudinal Survey of Youth with sibling fixed effects models to explore the effects of time receiving SNAP within disadvantaged neighborhoods on child obesity. Time participating in SNAP during ages 2–8 and ages 14–18 was associated with a lower proportion of time obese for children in disadvantaged neighborhoods, to an increasing degree as the level of neighborhood advantage declined. Given that most individuals who spend an extended period of time using SNAP live in the least advantaged neighborhoods, these results suggest that SNAP participation during these childhood years may help to reduce proportion of time obese as a child. Overall, results of this investigation suggest that participation in SNAP may have protective effects for children living in low-income households within disadvantaged neighborhoods.

KEYWORDS:

• Obesity
• SNAP
• neighborhood effects
• poverty
• BMI

Introduction

U.S. poverty programs have become arenas for behavior change, in some instances through the attachment of eligibility to socially desired behaviors and in other instances through the delivery of goods and services in ways that incentivize some behaviors and discourage others (Holtyn, Jarvis, and Silverman 2017). In the case of the Supplemental Nutrition Assistance Program (SNAP), initiatives currently under debate include applicant/recipient drug testing (NCSL 2017), work requirements (Brantley, Pillai, and Ku 2020; NCSL 2018), and increased restrictions to SNAP-eligible foods (NCSL 2019). Debates about these and other proposals are seeded not only by the theme of deservingness, but also by health-based concerns, further accented by projections that nearly half of adults in the U.S. will be obese by 2030 (Ward et al. 2019).

Proposals that directly or indirectly link program participation to unhealthy childhood weight persist despite inconsistent research evidence (Gundersen 2015; Kreider et al. 2012; Leung et al. 2017). Unpacking the assumptions that underlie such proposals requires tracing a path from family resources (in cash and vouchers) to purchasing decisions to children’s access and consumption, and only then to children’s weight. In one set of scenarios, SNAP allows families to shift cash spending away from food and toward items and activities that tend to increase the likelihood of obesity (e.g., electronic gaming) or decrease the likelihood of obesity (e.g., athletic leagues). Other, more direct hypothesized pathways rely on assumptions about whether an increase in food purchasing capacity will shift parents’ purchasing toward healthy or unhealthy options and/or serve to stabilize spending across time (Gundersen 2015).

In this study, we add two factors—age and place—to the potential pathways elaborated above. The neighborhoods in which families live and shop, and in which children play and eat, may influence purchasing options, eating behaviors, and recreational opportunities (DeBono, Ross, and Berrang-Ford 2012). Moreover, the roles of household food purchasing, as well as of neighborhood conditions and influences, are likely to change over the course of a childhood. For example, while young children eat most meals at home, school-age children are more likely to eat at school, where they consume lower amounts of low-nutrient, high-calorie foods (Briefel, Wilson, and Gleason 2009).

While Vartanian and Houser (2012) examined the effects of SNAP use and residential neighborhood during childhood on subsequent adult body mass index (BMI), we know of no work that examines the effects of SNAP use and residential neighborhood during childhood on childhood obesity at different childhood stages. Focusing on distinct periods of childhood (2–8, 9–13, 14–18), we used sibling fixed effects models to examine whether SNAP participation within varying neighborhood conditions was associated with the duration of time children were obese.

Background

SNAP participation

State and federal-level proposals for limiting SNAP access exist in the context of modest but steady multi-year declines in SNAP participation, SNAP spending, poverty, and food insecurity (defined by the U.S. Department of Agriculture [USDA] as “a household-level economic and social condition of limited or uncertain access to adequate food” [2019a, para. 5]). In fiscal year (FY) 2018, a monthly average of 40.4 million people, in 20.1 million U.S. households, participated in SNAP. These numbers represent a decline in participation from a historic high of 47.6 million persons in FY2013. As a proportion of the total population, participation has also declined, from 15.0% in 2013 to 12.3% in 2018. SNAP spending has also declined over the past five years, from approximately $79.9 billion in FY2013 to $65.3 billion in FY2018 (USDA 2019b).

Declines in SNAP participation have kept pace with declines in the U.S. poverty rate, which went from 14.8% in 2013 to 12.3% in 2017 (U.S. Census Bureau 2018). Food insecurity has also declined, with the proportion of U.S. households with food insecurity moving from 14.3% in 2013 to 11.8% in 2017 (USDA 2018).

SNAP participation and obesity

Against a backdrop of rising child and adult obesity rates in the United States (Hales et al. 2017; Ogden et al. 2015), researchers over the past two decades have explored whether SNAP participation increases individual risk of overweight or obesity. Studies have differed in multiple ways, including participant age, timing of conditions (cross-sectional, longitudinal), data source, and method (e.g., fixed effects, propensity score matching, difference-in-differences), and have yielded mixed results that are difficult to explain other than by reference to differences in the approaches used. In adults, SNAP participation has been linked to overweight and obesity for women but not for men (see reviews by DeBono, Ross, and Berrang-Ford 2012; Gundersen 2015; Ver Ploeg and Ralston 2008). By contrast, two more recent studies—one of SNAP participation in food insecure households (Nguyen et al. 2015) and another of SNAP benefit increases in households with young children (Almada and Tchernis 2018)—found SNAP participation to have null or negative effects on BMI and obesity risk for adults in these households. Overall, studies of children and adolescents have found small or no relationships between SNAP use and obesity (Bhattacharya and Currie 2001; Gibson 2004; Kreider et al. 2012; Leung et al. 2013; Schmeiser 2012; Simmons et al. 2012). However, based on a systematic review of studies of SNAP participation and weight outcomes in children and adolescents, Hudak and Racine (2019) observed that a subset of studies addressing selection bias observed positive associations between SNAP use and overweight, particularly for girls.

Overweight and obesity during childhood are of particular concern in part because children who are overweight and obese tend to maintain these weights as adults (Simmonds et al. 2016). Obesity has been associated with a number of health problems, including high blood pressure, high cholesterol, and several types of cancers (Freedman et al. 2007; Lauby-Secretan et al. 2016), as well as with increased psychological morbidity and psychosocial risk in children and adolescents (Rankin et al. 2016).

Relationship mechanisms

Theorizing about a positive relationship between household SNAP receipt and overweight/obesity in children generally rests on the premise that, relative to SNAP non-recipients, recipients either (a) purchase more food overall, leading to greater food access and consumption beyond what is needed for children; (b) purchase greater proportions of unhealthy than healthy foods, leading to children consuming less nutritious, more calorie-dense foods; or (c) shift what would otherwise have been food spending toward items or activities that produce more sedentary behaviors in children (Gundersen 2015; Gundersen 2016). Overall, the same or similar theories have been used to explain relationships between SNAP use and health outcomes in adults, adolescents, and children, although, as we explore further below, there are a number of reasons why the pathways between SNAP use and overweight may be different for children than for adults, as well as for children at different childhood stages.

According to Ver Ploeg and Ralston (2008), one of the primary explanations for a SNAP and obesity connection is the “food stamp cycle,” the practice of distributing SNAP funds such that temporary periods of abundance are followed by periods of scarcity (Council of Economic Advisors 2015). Thus, recipients and their children are thrown into a “feast and famine” cycle, which researchers have found to have positive effects on BMI for adults (Shapiro 2005; Smith, Stillman, and Craig 2017; Yanovski 2003). Another primary explanation, the “income effect,” stems from the inelasticity of SNAP income, which can be used exclusively to purchase food. In this case, low-income individuals purchase and consume more food, or less healthy food, than they otherwise would, increasing their risk of overweight/obesity (Fox, Hamilton, and Lin 2004; Leung and Villamor 2011).

However, also plausible is the possibility that SNAP income allows households to shift spending toward foods that are healthier and, in some cases, more expensive (Darmon, Ferguson, and Briend 2003; Hastings, Kessler, and Shapiro 2019; Kern et al. 2017) or even toward physical activities with weight-reducing effects (Gundersen 2015). If SNAP participation is related to lower likelihoods of overweight/obesity, another mechanism might be through reductions in food insecurity, which itself has been linked to child obesity (Casey et al. 2006).

Research on children and adolescents

With some exceptions, studies of SNAP participation among children and adolescents have found small or no associations between program participation and BMI, overweight, or obesity. Using child fixed effects models and data from the National Longitudinal Survey of Youth (NLSY79), Gibson (2004) concluded that long-term food stamp use was positively associated with overweight in girls ages 5–11, but negatively associated with overweight in boys in the same age range. By contrast, using the same dataset with an instrumental variables approach, Schmeiser (2012) found that SNAP participation reduced BMI percentile and the likelihood of overweight or obesity for boys and girls ages 5–11, relative to children of the same ages in SNAP-eligible but non-participating households. Using a sample of children in Head Start, Simmons et al. (2012) observed no differences in BMI percentile for children from SNAP-recipient and non-recipient households. In a study that addressed selection and misreporting bias using partial identification and bounding methods, Kreider et al. (2012) found that SNAP participation reduced the probability of obesity among children ages 2–17 when misreporting rates were less than 4%.

Relatively few studies have focused on adolescents, either alone or as a subgroup of a larger sample. Bhattacharya and Currie (2001) and Gibson (2004) found no relationship between food stamp use and obesity for adolescents ages 12–16 and adolescents ages 12–18 respectively. By contrast, Schmeiser (2012) concluded that SNAP participation reduced BMI/obesity risk for boys ages 12–18 with no effect for girls in the same age bracket. Based on a systematic review of 23 studies of children ages 2–18, Hudak and Racine (2019) concluded that the accumulated evidence supports either no effect or a beneficial effect of SNAP participation on healthy weight for boys, while, for girls, participation may add to the risk of overweight and obesity, particularly if their participation is long-term.

The primary mechanisms posited for a SNAP-obesity relationship in the studies above—namely, the food stamp cycle and income effects—may function differently in different places (e.g., by food store availability, transportation access, food prices, or access to recreational activities). Below, we describe studies connecting neighborhood domains to weight outcomes and then explore how these domains together might influence a SNAP-obesity connection in different ways for preschool and primary-school-age children, middle-school-age children, and adolescents.

Neighborhoods and obesity

Studies connecting residential neighborhoods to child and adult BMI, overweight, and obesity have produced mixed results, perhaps not surprisingly given challenges attached to determining what geographic, social, or other indicators constitute a given individual’s “neighborhood” (Diez Roux and Mair 2010). According to Carroll-Scott et al. (2013), neighborhood environments have three key domains: socioeconomic, built, and social. The domain that we explicitly examine in this study—socioeconomic environment—includes the individual and composite socioeconomic characteristics of those residing in a neighborhood. The built environment pertains to the way space is designed and used, and the social environment “ … refers to relationships, groups, and social processes that exist between individuals and groups who live and work in a neighborhood” (Carroll-Scott et al. 2013, 107). Because studies of built and social environments require information on attributes that are rarely systematically and uniformly documented across the United States (e.g., local food store inventory and pricing, availability of public parks and recreation centers, social networks and ties), longitudinal and nationally representative studies tend to rely on indicators of the socioeconomic environment.

Socioeconomic environment

Previous studies of the socioeconomic environment have linked neighborhood socioeconomic conditions to weight outcomes, including obesity risk (Grow et al. 2010) and BMI scores (Burdette and Needham 2012). Kowaleski-Jones and Wen (2013), found that, as the percentage of below-poverty-level households in the census tract increased, so did children’s risk for being overweight. In a study using National Longitudinal Study of Adolescent Health data, Lippert (2016) observed that adolescents who lived in poor neighborhoods had higher obesity risks in adulthood than those who did not, but that moving out of high-poverty conditions during the transition to adulthood lessened this risk. In a study that looked at SNAP use in neighborhood contexts, Vartanian and Houser (2012) concluded that SNAP use during childhood increased individuals’ risk of adult obesity overall. However, this was particularly true for those individuals who grew up receiving SNAP in neighborhoods that were more advantaged; by contrast, those who had grown up in disadvantaged neighborhoods had fairly stable adult BMI predictions regardless of the extent of childhood SNAP use.

Several studies have used the racial composition of a neighborhood (i.e., proportions of White or Black residents) in models of socioeconomic environments. Neighborhood racial composition (often as an indicator of residential segregation or environmental racism) has been shown to be related to food access, both overall (Powell et al. 2007), and in interaction with neighborhood poverty (Zenk et al. 2005). For example, in a study of Detroit, Michigan, Zenk et al. (2005) reported that the lowest income Black neighborhoods in the sample were located 1.1 miles farther from a supermarket than the lowest income White neighborhoods.

Built environment

Aspects of a neighborhood’s built environment have been hypothesized to influence obesity risk both directly and indirectly via food access, dietary quality, safety, and/or outlets for physical activity. Attributes including access to “healthy” food outlets (Carroll-Scott et al. 2013; Jennings et al. 2011) and lower food prices, especially prices for fruits and vegetables (Morrissey, Jacknowitz, and Vinopal 2012; Powell and Bao 2009; Sturm and Datar 2005), have been linked to reduced risk of overweight and lower BMI scores in children. Moreover, young adults at the lower ends of the income and education spectrum have been observed to be more price-sensitive than others in their consumption of fruits and vegetables (Powell, Zhao, and Wang 2009). By contrast, Ford and Dzewaltowski (2011) and Drewnowski et al. (2012) found that supermarket availability and distance from home had no influence on BMI in samples of adults.

Studies linking children’s BMI/obesity to characteristics of the built environment relating to safety and outlets for physical activity have also yielded mixed results. Poor housing and lack of sidewalks, parks, and recreation centers have been associated with higher odds of being overweight or obese in children (Fan and Jin 2014; Singh, Siahpush, and Kogan 2010). However, in a national sample of children ages 2–11, Kowaleski-Jones and Wen (2013) found no relationship between the amount of local green space or proximity to parks and risk of overweight.

Based on a systematic review of studies relating local food environments to obesity, Cobb et al. (2015) concluded that most reported associations were null and that study quality was, for the most part, low. Similarly, a systematic review of geospatial analyses found that while most (80%) of the studies reviewed identified at least one statistically significant association between local food environments and obesity, variations in the ways that researchers identify, map, and group food sources make interpretation and application of results difficult (Gamba et al. 2014). According to Adachi-Mejia et al. (2017), another explanation for inconsistent results reported across studies to date may lie in differences between urban and non-urban settings, as well as between geographic regions.

Social environment

The neighborhood social environment may influence food purchase, food consumption, physical activity, and body weight via a number of mechanisms, including social modeling, social ties, social capital, and collective action (Carroll-Scott et al. 2013). Based on data from a natural experiment involving adolescents (ages 16–19) from military families, Datar, Mahler, and Nicosia (2020) found that living in a county with an above-median obesity rate increased adolescents’ likelihood of selecting an overweight or obese body type as their ideal, as well as their individual likelihood of being overweight or obese. In a study comparing the physical and social aspects of children’s environments, Franzini et al. (2009) observed that the social, and not the physical, environment positively influenced children’s physical activity and that, in turn, physical activity was negatively associated with obesity. A 2014 review of studies connecting social modeling to food choice or consumption concluded that the accumulated evidence, while mixed, is particularly strong for the influence of those with whom individuals want to associate and those whom individuals perceive as similar to themselves (e.g., by race, ethnicity, and social class; Cruwys, Bevelander, and Hermans 2014). Cannuscio et al. (2014) focused on health-related food shopping in a single U.S. city and observed that perceived similarities in identity to other customers, positive interactions with employees and other customers, and several aspects of physical environment influenced participants’ choices on where to purchase food.

Notably, the three neighborhood domains described above are interrelated and, as such, often difficult to disentangle. For example, when Cannuscio et al. (2014) identified social factors influencing food shopping choices, they included characteristics that are also often identified as aspects of built environments, including a store’s safety and accessibility. They also noted that individuals demonstrated preferences for shopping with those perceived to be similar in ways related to neighborhood socioeconomic environments, including income, education, and race.

SNAP participation and body weight: by place and age

The ways that household SNAP participation and neighborhood conditions interact to influence body weight may be different for children than for adults and may vary by childhood stage. As previously discussed, one of the premises underlying the food stamp cycle is that the insufficiency of benefits to cover the full month will lead families to have periods of excess consumption (“feast”) followed by periods of under-consumption (“famine”). However, children who are in school, sometimes for two full meals per day, may be less susceptible to this cycle than those who are either too young for full-day school or old enough to leave school for periods of the day. Indeed, most vulnerable to the food stamp cycle might be pre-school-aged children in neighborhoods that are both urban and low-income, where food costs tend to be higher than in suburban and higher-income areas (Hendrickson, Smith, and Eikenberry 2006; Kern et al. 2017; Powell et al. 2007), increasing the likelihood that benefits will be insufficient for the month.

The income effect posits that low-income individuals who receive SNAP may purchase more food overall, or greater proportions of unhealthy food, than low-income individuals who do not receive SNAP. For example, in a study of adults residing in low-income census tracts in Massachusetts, Webb et al. (2008) found that SNAP participation was associated with higher BMI independent of household-level food insecurity. The income effect may exert a stronger influence on adolescents than on younger children to the extent that their social models (often peers; Salvy et al. 2012) engage in similar patterns of purchasing and consumption (Cannuscio et al. 2014).

As noted by Gundersen (2015), it is also possible that SNAP income allows families to shift spending toward healthier foods or health-promoting, weight-reducing activities. As suggested above, the possibility of doing so depends in part on the availability and accessibility of such foods and activities in the local area and/or on one’s potential to travel beyond the local area. Families who are geographically constrained, as many low-income families are (Shannon and Christian 2017), may both shop and recreate locally, in the advantaged or disadvantaged areas in which they live, making the additional purchasing power conferred by SNAP receipt especially important for those in disadvantaged areas where availability and access tend to be more limited (Walker, Keane, and Burke 2010; Zenk et al. 2011). Kim (2016) addressed this possibility directly, reporting that the 2009 SNAP benefits increase resulted not only in increases in food expenditures but also in spending increases for things like transportation, housing, and education.

For young children in particular, food insecurity has been linked directly to neighborhood poverty, even after individual family characteristics have been controlled (Morrissey et al. 2016). If SNAP receipt reduces children’s food insecurity (Collins and Klerman 2016), we can, in turn, anticipate a reduction in the likelihood of obesity for young children (Casey et al. 2006). Adults with young children may be constrained from traveling to stores with lower food prices because of the potential difficulties of traveling with a young child. These constraints become greater when families do not have cars or access to affordable public transportation or childcare (Hendrickson, Smith, and Eikenberry 2006).

To determine how children at various stages of childhood are affected by SNAP participation and neighborhood environment, we examined age brackets that correspond roughly to preschool and primary school age (ages 2–8 years); middle school age (ages 9–13 years); and adolescence (ages 14–18 years). Several past studies have employed these childhood stages and found statistically significant relationships between childhood conditions, including neighborhood advantage and parental income, and child or adult outcomes, including completed schooling and adult income (Duncan et al. 1998; Levy and Duncan 2000; Vartanian and Buck 2005).

Hypothesis

Focusing on three distinct periods of childhood, we examined whether SNAP participation within disadvantaged neighborhoods was associated with the proportion of time children’s body weights were above the 95th percentile for their age and gender (i.e., obese).

Data and methods

Samples

The sample included U.S.-resident children aged 2–18, who were born between 1986 and 2010, and whose mothers participated in the 1986–2012 Child and Young Adult sample of the National Longitudinal Survey of Youth (NLSY79: CYA). Collected annually from 1979 to 1994 and biennially from 1996 to 2012, the NLSY79 followed youth born between 1957 and 1965, as they grew into adulthood and formed families of their own. Collected biennially from 1986 to 2012, the CYA sample followed children born to the women of the NLSY79. Because the NLSY79 over-sampled Black, Hispanic, and low-income non-Black, non-Hispanic individuals, sampling weights were used to make the sample representative of both U.S. women born between 1957 and 1965 and their children.

In addition to information on individual and family characteristics from both childhood and early adult years, the NLSY79: CYA reports children’s height and weight information for each of the biennial survey years. Using a SAS program for statistical analysis created by the Centers for Disease Control and Prevention (CDC), we converted children’s heights and weights to BMI values and their age- and gender-specific percentiles for those aged 2–18 years.

We merged the NLSY79: CYA with data from the U.S. Census Bureau in order to obtain neighborhood information at the level of census tracts. For data years 1986–1994, we used 1990 Census information; for 1996–2004, we used 2000 Census information; and for 2006–2012, we used information from the 2006–2010 American Community Survey, which gives average census tract information over this five-year period. To illustrate the linkage between neighborhood (i.e., census tract) and child data over time, if a child lived in census tract A in 1986, her individual-level record for that year would be connected to 1990 U.S. Census tract-level data for census tract A. If that same child moved to census tract B in 1988, her record for that year would be connected to 1990 U.S. Census tract-level data for census tract B. We have examined other approaches to linking Census and national survey data but found these alternative approaches to make little difference to the results (Vartanian and Houser 2012).

To allow for the use of sibling fixed effects models, we constructed two samples: one for all children between the ages of 2 and 18 (hereafter, full sample; n = 8,620), and one for all children who have siblings in the sample (hereafter, siblings-only sample; n = 7,756).

Dependent variable

The dependent variable was the proportion of time within a given childhood stage (2–8, 9–13, 14–18) spent obese (childhood BMI percentile above the 95th percentile, indexed to age and gender). Measuring the dependent variable as a proportion of time, rather than a point in time, makes for a more nuanced and potentially more meaningful indicator, given evidence that the health effects of obesity tend to be sensitive to both timing and duration (The, Richardson, and Gordon-Larsen 2013).

Independent variables

Program participation and household income

We constructed the primary set of independent variables based on both poverty program participation and household income relative to the Federal Poverty Line (FPL). For each childhood stage (2–8, 9–13, 14–18), variables were created for the proportions of time spent in each of the following conditions:

• participating only in SNAP or the Food Stamp Program (hereafter, SNAP);

• participating only in Temporary Assistance for Needy Families (TANF) or Aid to Families with Dependent Children (hereafter, TANF);

• participating in both SNAP and TANF;

• household income below 150% of FPL (“in poverty”); participating in neither SNAP nor TANF;

• household income between 150% and 200% of FPL (“near poverty”); and

• household incomes above 200% of FPL (“not in poverty”).

The focal comparison for this study was between the proportion of time spent using only SNAP and the proportion of time spent in poverty, with neither SNAP nor TANF participation.

Across U.S. states, income maximums for SNAP eligibility have been generally higher than those for TANF eligibility, while restrictions on SNAP participation (e.g., work effort requirements) have been fewer than those for TANF participation. Together, these factors mean that overlap in program participation for households is far from 100%, with rates of SNAP participation far exceeding those of TANF participation (U.S. Bureau of Labor Statistics 2018).

In order to determine program participation and income during each survey year, we examined each month to determine (1) participation in SNAP, TANF, both programs, or neither program, and (2) the child’s family’s household income. We then calculated the proportions of time children spent in each program and income category per year. For example, if the child’s family spent six months receiving SNAP, spent no months receiving TANF, and had yearly income below 150% of the FPL, the family would be coded as having spent 50% of time with SNAP-only and 50% of time in poverty with neither TANF nor SNAP. Annual proportions were then averaged across survey years within each childhood stage.

Neighborhood conditions

The neighborhood conditions measure was constructed through principal component analysis (PCA) using indicators for income, employment, and racial composition, following the work of previous researchers (Wodtke, Harding, and Elwert 2011; Vartanian and Houser 2012). Specifically, the neighborhood conditions index included the census tract’s poverty rate, severe poverty rate (i.e., incomes below half the poverty line), employment and unemployment rates by gender, percentage of women-headed households, percentage receiving public assistance income, percentage of White and Black households, and three income measures (all in 2011 dollars): percentages of households with incomes under $15,000, between $60,000 and $125,000, and above $125,000. Generating index values through PCA allowed us to include multiple neighborhood indicators in the analyses without introducing collinearity threats.

Higher index values indicate more advantaged neighborhoods. The variable is centered on its mean such that a score of 0 indicates an “average” neighborhood and scores of 1 and –1 indicate neighborhoods one standard deviation (SD) above and below the mean, respectively.

Table 1 shows each indicator comprising the neighborhood conditions index, its loading, and the correlation between that indicator (i.e., the original neighborhood variable) and the principal component (PC) variable. Each component is highly correlated with the PC variable, with the lowest absolute value of a correlation at .68 for both the full and siblings-only samples and most of the correlations at or above .80. The principal component variable explains about 67% of the variance in the set of original variables for both samples.

Table 1. Neighborhood index: indicators, loadings, and correlations (r).

	Full		Siblings
	Eigenvector	PC, r	Eigenvector	PC, r
% below poverty level	–.31	–0.96	–.31	–0.96
% below half of poverty level	–.30	–0.93	–.30	–0.93
% women unemployed	–.28	–0.86	–.28	–0.86
% men unemployed	–.28	–0.85	–.28	–0.86
% women employed	.23	0.71	.23	0.71
% men employed	.28	0.85	.28	0.85
% women–headed families	–.29	–0.88	–.29	–0.88
% receiving public assistance	–.30	–0.91	–.30	–0.91
% White	.23	0.70	.23	0.70
% Black	–.22	–0.68	–.22	0.68
% with incomes <$15k	–.31	–0.95	–.31	–0.95
% with incomes $60k–$125K	.29	0.88	.29	0.89
% with incomes >$125K	.25	0.75	.25	0.75
% variation explained	66.60		67.04
n	8,620		7,756

We interacted the neighborhood conditions variable with the SNAP-only program participation variable to model the relationship between SNAP use and childhood time spent obese in varied neighborhood contexts during particular childhood stages. As previously noted, our focus was whether, in three distinct periods of childhood, receipt of SNAP benefits in the context of disadvantaged neighborhoods was associated with the proportion of time children were obese.

Control variables

We examined previous studies on neighborhood effects and SNAP effects as well as our hypotheses to determine the appropriate control variables for the study (Gibson 2011; Kowaleski-Jones et al. 2017; Vartanian and Houser 2012). Control variables for all models included the following, most averaged over the particular childhood stages: child’s mother variables (BMI, age, marital status, years of education, and any work limits); family variables (number of children in the household, age of the youngest child, income relative to the poverty line [a linear measure of income, to supplement the non-linear independent variables], whether the house is owned, value of the house if owned [as an assets indicator; Drewnowski et al. 2015], whether the family owns a car, hours the parents work, whether the family is living in the south [where child obesity rates are the highest relative to other U.S. regions, according to the U.S. Department of Health and Human Services in 2017], and whether the family is living in a rural area relative to an urban area); and child variables (gender and the year the child started in the sample). We also included state TANF maximums for a family of four.

Statistical methods

We examined relationships between SNAP participation, neighborhood conditions, and obesity in childhood using several modeling strategies: Ordinary Least Squares (OLS) regression with the full sample, OLS regression with the siblings-only sample, and sibling fixed effects (FE) regressions. Because the OLS model results for both the full and siblings-only samples were highly similar, only the full-sample OLS and siblings-only sample FE models are shown.

Sibling Fixed Effects models

Self-selection is a well-documented source of bias in estimates of program participation effects as well as neighborhood conditions (Diez Roux and Mair 2010; Yen et al. 2008). People “select” where they live and whether they participate in programs like TANF and SNAP for myriad reasons, which, if either unobservable or not captured by control variables, may be mistaken for neighborhood or program participation effects. Such reasons may include family norms and traits. We addressed these types of unobservable and shared attributes within families by using sibling FE models (Aaronson 1998). Sibling FE models compare siblings from multi-child families to each other, while holding unobservable and unvarying family background factors constant.

To estimate effects in FE models, there must be variation among siblings. For neighborhood variables, variation among siblings comes from changing conditions in the same neighborhood over different age periods for different siblings, from family moves into different neighborhoods at different sibling age periods, and/or from one child being in the family in a particular time period while the other is not (e.g., because a second child has not yet been born or is no longer living with the family). We found large within-family differences for key study variables including the proportion of childhood time spent obese (72% of families), proportion of time in poverty (67% of families), and the neighborhood index (95% of families). Almost 38% of families had differences in the proportion of time spent with SNAP.

Sibling FE models have four primary limitations. First, they require that children have siblings. Consequently, sample sizes for sibling FE models are smaller than those for models that use all children within the sample. Second, such models require a dummy variable for each family (except the reference group family), using up many degrees of freedom. Third, FE models control only for those unobservable factors that are unvarying among the siblings, such as fixed characteristics of their parents or shared genetic makeup. Fourth, if children from single-child families are more or less prone to obesity than those from multi-child families, sibling model results will not be generalizable beyond children with siblings. To help determine if the first limitation affected our results, we compared the siblings-only and full samples both descriptively and by running OLS regression analyses on both.

In order to examine the independent effects of each of the neighborhood conditions, we also ran separate models for each neighborhood condition along with all control variables.

Results

Descriptives

Table 2 compares the average childhood characteristics for both the full and siblings-only samples, for children in each childhood stage. Differences between the two samples were consistently small. For example, among children ages 2–8, the mean proportion of time obese is 14% for the full sample and 14% for the siblings-only sample.

Table 2. Weighted mean values for focal study variables.

	Ages 2–8		Ages 9–13		Ages 14–18
	Full	Siblings	Full	Siblings	Full	Siblings
Proportion of time obese	.14 (.29)	.14 (.15)	.14 (.33)	.14 (.33)	.12 (.31)	.12 (.31)
Proportion of time on SNAP	.04 (.17)	.04 (.19)	.04 (.19)	.05 (.20)	.04 (.19)	.05 (.20)
Proportion of time on TANF/SNAP	.08 (.27)	.07 (.27)	.09 (.29)	.08 (.28)	.04 (.20)	.04 (.21)
Proportion of time in poverty	.18 (.28)	.20 (.33)	.21 (.32)	.20 (.33)	.19 (.34)	.20 (.34)
Proportion of time near poverty	.10 (.18)	.10 (.21)	.11 (.21)	.10 (.21)	.09 (.21)	.09 (.21)
Proportion of time not in poverty	.58 (.43)	.58 (.45)	.54 (.44)	.56 (.44)	.63 (.45)	.61 (.45)
n	7,505	6,509	6,413	5,389	5,843	4,862

Table 3 describes the average neighborhood conditions index scores (as well as the 25th and 75^th percentile scores) for six sub-groups of children: those who spent some (more than 25%) of their childhood time on SNAP (n = 816), those who spent most (more than 50%) of their childhood time on SNAP (n = 254), those who spent some of their childhood time in poverty without TANF/SNAP assistance (n = 2,932), those who spent most of their childhood time in poverty without TANF/SNAP assistance (n = 1,232), those who spent some of their childhood time not in poverty (i.e., with household incomes above 200% of FPL; n = 4,461), and those who spent most of their childhood time not in poverty (n = 3,538). Average neighborhood index scores were highest (.56) for those who spent most of their childhoods not in poverty and lowest (-.92) for those who spent most of their childhoods on SNAP. Among children in poverty, those who received SNAP tended to live in less advantaged neighborhoods than those who did not receive SNAP.

Table 3. Neighborhood conditions index scores, by program participation and household income conditions.

	M (SD)	25th percentile	75th percentile	n
% time on SNAP>25%	–.76 (1.00)	–1.31	–.05	816
% time on SNAP>50%	–.92 (1.16)	–1.43	–.16	254
% time in poverty without TANF/SNAP>25%	–.28 (.87)	–.85	.34	2,932
% time in poverty without TANF/SNAP>50%	–.33 (.82)	–.88	.27	1,232
% time not in poverty>25%	.46 (.75)	.09	1.00	4,461
% time not in poverty>50%	.56 (.68)	.23	1.06	3,538

Note: Neighborhood index values are standardized (mean = 0; SD = 1).

Regression results

Table 4 shows OLS regression and FE model results for the siblings-only sample. Relationships between the independent variables (neighborhood conditions and proportion of childhood time with SNAP) and the dependent variable (proportion of childhood time obese) are presented for each childhood stage. Nearly identical results for the full (results not shown) and the siblings-only samples indicated that the results were not driven by examining siblings; thus, we reported results only for the siblings-only sample. (Full results available upon request).

Table 4. OLS and FE regression results for childhood proportion of time obese.

	Ages 2–8		Ages 9–13		Ages 14–18
Variable	Siblings, OLS	Siblings, FE	Siblings, OLS	Siblings, FE	Siblings, OLS	Siblings, FE
Neighborhood conditions index	–.02 (.01)***	–.01 (.01)	–.01 (.01)	.00 (.01)	–.01 (.01)*	–.01 (.01)
Neighborhood conditions index* % time on SNAP	.05 (.02)*	.09 (.04)*	.03 (.02)	.02 (.03)	.04 (.02)^+	.05 (.03)^+
% time on SNAP	.04 (.03)	.04 (.05)	.00 (.03)	–.00 (.05)	.02 (.03)	.07 (.05)
n	6,509	6,509	5,389	5,389	4,862	4,862
Within R^2 /Adjusted R^2	.08	.06	.10	.04	.09	.01
# of families	2,481		2,058		1,888

Note: Models include all control variables. +:p < .10;*:p < = .05; **:p < = .01; ***:p < = .001.

For those in every childhood stage, as well as in both OLS and sibling FE models, SNAP exposure had no direct association with obesity. However, for children in the youngest and the oldest age cohorts, respectively, we found a relationship between the SNAP exposure/ neighborhood conditions interaction and the proportion of time obese, with FE model coefficients of .09 for the 2–8 year olds (p = .02) and .05 for the 14–18 year olds (p = .08).

For 2–8 year olds, the coefficient for the FE model interaction indicates that those who received SNAP for their entire 2–8 age period and who lived in neighborhoods that were one SD above the mean are predicted to spend 9 percentage points more time obese as a child, relative to those who were in poverty without TANF/SNAP assistance for the entire period. Conversely, those who received SNAP for their entire 2–8 age period and lived in neighborhoods that were one SD below the mean are predicted to spend 9 percentage points less time obese, relative to those who were in poverty without TANF/SNAP assistance for the entire period. As neighborhood index scores decline (i.e., neighborhoods become less advantaged), the relationship between SNAP exposure and time spent obese becomes larger, with children in SNAP-recipient households in poverty spending less time obese than those in non-SNAP recipient households in poverty. Similar and marginally significant results were observed for the 14–18 year-olds, albeit at around half the size of the 2–8 year-olds. For the 9–13 year-olds, however, there was no relationship between SNAP exposure, neighborhood index score, and time spent obese.

Because most children who live in SNAP-recipient households for large proportions of their childhoods also live in less advantaged neighborhoods (see Table 3), the regression results suggest that receiving SNAP for a longer period of time will, on average, decrease proportion of time obese for children in the 2–8 and 14–18 age groups. To further illustrate this point, we examined the distribution of children who spent some (more than 25%), or most (more than 50%), of their time on SNAP across various neighborhood conditions index scores. More than three of every four children who spent some or most of their childhoods with SNAP (77% and 80% respectively) lived in neighborhoods below the index mean, with over 36% and 41% respectively living in neighborhoods over 1 SD unit below the mean (“less advantaged”). Conversely, only 0.5% of those who spent some of their childhoods with SNAP, and none of those who spent most of their childhood times with SNAP, lived in neighborhoods 1 SD unit above the mean (“more advantaged”).

To illustrate the direct effects of each of the 13 neighborhood characteristics comprising the neighborhood conditions index, Table 5 shows results of FE models regressing childhood time spent obese on each neighborhood condition, the proportion of childhood time receiving SNAP, and the interaction between the neighborhood condition and the proportion of childhood time receiving SNAP. Only estimates that were statistically significant for time using SNAP, the neighborhood condition, or the interaction of these two are shown, reducing the original 13 characteristics to 11. The racial segregation indicators were the two indicators not related to the proportion of time children were obese. As detailed on Table 5, results for discrete components of the neighborhood conditions index were similar to and in the same direction as those for the index overall.

Table 5. Proportion of childhood time spent obese: fixed effect models.

Neighborhood and SNAP variables	Ages 2–8	Ages 9–13	Ages 14–18
Poverty rate	.02 (.01)		.01 (.01)
Poverty rate*time on SNAP	–.12 (.04)***		–.05 (.02)^+
Time on SNAP	.08 (.05)		.06 (.04)
% Public assistance		.01 (.01)
% Public assistance*time on SNAP		–.05 (.03)^+
Time on SNAP		.01 (.04)
Severe poverty rate	.01 (.01)		.01 (.01)
Severe poverty rate*time on SNAP	–.06 (.04)^+		–.04 (.01)^+
Time on SNAP	.01 (.05)		.06 (.04)
% with female heads			.02 (.01)
% with female heads*time on SNAP			–.04 (.03)^+
Time on SNAP			.05 (.04)
Female unemployment rate			.03 (.01)**
Female unemployment rate* time on SNAP			–.03 (.03)
Time on SNAP			.05 (.05)
Female employment rate	–.01 (.02)	.02 (.02)
Female employment rate* time on SNAP	.05 (.03)^+	.06 (.04)^+
Time on SNAP	–.01 (.05)	–.08 (.06)
Male unemployment rate	.03 (.01)*		.02 (.01)^+
Male unemployment rate* time on SNAP	–.04 (.04)		–.01 (.02)
Time on SNAP	–.01 (.06)		.04 (.04)
Male employment rate	.00 (.01)
Male employment rate* time on SNAP	.06 (.03)*
Time on SNAP	–.00 (.05)
% w/income<$15,000	.00 (.01)		–.01 (.01)
% w/income<$15,000*time on SNAP	–.08 (.03)*		–.07 (.04)^+
Time on SNAP	.04 (.06)		.07 (.05)
% w/$60,000<income<$125,000	–.03 (.01)*		–.02 (.01)^+
% w/$60,000<income<$125,000*time on SNAP	.15 (.05)*		.05 (.04)
Time on SNAP	.09 (.06)		.07 (.05)
% w/income>$125,000	–.04 (.02)***
% w/income<$125,000*time on SNAP	.20 (.06)***
Time on SNAP	.09 (.06)
n	6,509	5,389	4,862

Note: Models include all control variables. Each neighborhood characteristic is run in a separate model and only statistically significant coefficients are shown. All neighborhood characteristics are standardized (mean = 0; SD = 1).

+:p < .10; *:p < = .05; **:p < = .01; ***:p < = .001.

Discussion, limitations, and conclusion

State and federal governments have taken steps to try to reduce the obesity rate for children in the United States. Proposals targeting the SNAP program often invoke rising obesity rates as justifications for program changes. For example, policies that propose limits on the types of foods that can be purchased via SNAP (e.g., limiting purchasing of sugar-sweetened beverages) could make SNAP more effective at enabling healthier diets for families, thereby reducing the risk of diet-related diseases in children. To be able to make effective policy recommendations for SNAP, we need to understand if and when exposure to SNAP in childhood leads to poor nutritional health outcomes, like obesity risk.

This study posed the question of whether SNAP usage and neighborhood conditions during childhood were associated with the proportion of time children were obese. In every model and for every childhood stage, SNAP participation had no direct effect on obesity. Importantly, however, time receiving SNAP benefits during ages 2–8 and 14–18 reduced proportion of time obese for those living in less advantaged neighborhoods. As our descriptive results suggest, most people who receive SNAP for an extended period in childhood live in less advantaged neighborhoods. SNAP enrollment may act as a protective factor against obesity for children in these neighborhoods.

To the existing literature on SNAP participation and childhood obesity, our study adds a clarification of differences by duration of receipt, childhood stage, and neighborhood conditions. If we look only at the direct effects of SNAP, our study results are similar to those of Simmons et al. (2012), who found no differences in BMI percentile for preschool-aged children from SNAP-recipient and non-recipient households. However, in interaction with neighborhood conditions and for children in less advantaged neighborhoods, our results are consistent with those of Schmeiser (2012) and Kreider et al. (2012), who observed that SNAP participation reduced the likelihood of obesity in children.

The finding of no direct relationship between SNAP participation and childhood obesity for adolescents is consistent with evidence from Bhattacharya and Currie (2001) and Gibson (2004). Although Schmeiser (2012) did not account for neighborhood conditions, Schmeiser’s observation of a negative effect of SNAP participation on BMI/obesity for adolescent boys aligns with our finding of a negative effect for those adolescents who live in less advantaged neighborhoods.

Of the theories used to explain relationships between food assistance programs and obesity, our study results are consistent with the possibility that SNAP income allows households to shift spending toward healthier, more expensive foods or toward physical activities with weight-reducing effects (Gundersen 2015). In light of some evidence that healthy foods are pricier in low-income neighborhoods (Walker, Keane, and Burke 2010), when SNAP gives low-income families the opportunity to afford these foods, their purchasing decisions become less like (and of better nutritional quality than) the purchasing decisions of low-income families without SNAP. It is similarly possible that SNAP income reduces food insecurity which in turn lowers children’s obesity risk (Casey et al. 2006). SNAP may serve as a means for erasing the gap in consistent access to affordable and healthy food for low-income families and, thus, to a healthy weight for those who live in less advantaged areas.

As suggested by our results, these mechanisms may be amplified for families with young children (ages 2–8), because the constraints that are introduced by limited access to transportation and affordable child care in less advantaged neighborhoods (Hendrickson, Smith, and Eikenberry 2006) make parents more likely to shop within these local neighborhoods. Another explanation for the finding that SNAP receipt reduces the proportion of time obese for children ages 2–8 in less advantaged neighborhoods emerges from studies of food insecurity. According to Morrissey et al. (2016), neighborhood poverty is related specifically to child-level food insecurity for young children. If, as suggested by Collins and Klerman (2016), SNAP receipt reduces children’s food insecurity, we can anticipate a reduction in obesity for young children (Casey et al. 2006) in less advantaged neighborhoods.

Previous research offers few clues to the finding of a marginally significant relationship between SNAP receipt and proportion of time obese for 14–18 year olds in less advantaged neighborhoods. One possibility is that compared to adolescents from low-income households and less advantaged neighborhoods who do not receive SNAP, adolescents in low-income households and less advantaged neighborhoods that do receive SNAP derive particular benefits from the freeing up of household income (Kim 2016) for physical activities, athletics, and clubs. In the United States, ages 14–18 correspond to high school years, when the number of options for extracurricular activities tend to increase, and individuals acquire more control over where and how to spend their time and resources. Under these circumstances, household SNAP participation may simultaneously free up household income to purchase healthy food and for physical activities, both contributing to reduced likelihoods of obesity (albeit not to the extent we see in children ages 2–8).

The fact that for those in less advantaged neighborhoods, we observed protective effects for SNAP on obesity for the youngest (2–8) and (to a lesser extent) oldest (14–18) age cohorts, but not for those in between (9–13), is difficult to explain. For those age 9–13, regardless of neighborhood conditions, null results indicate that receiving SNAP neither hurts nor helps the proportion of time obese. In the United States, ages 9–13 correspond to the onset of puberty, which itself has been shown to have a complex, correlational relationship with overweight and obesity (Li et al. 2017). Future studies with larger samples may be able to identify with greater precision where and why differences in effect by age appear.

In addition to the limitations attached to the use of sibling FE models described above, our study was also limited by the fact that the neighborhood index does not contain characteristics for which we did not have data access (e.g., food prices, availability of clubs and activities). Moreover, we are unable to distinguish between children who spend most or all of their time in their residential neighborhoods and those who may be spending substantial time elsewhere, such as in the households of family members or friends in other census tracts. Another limitation pertains to the age brackets we selected; using different age brackets may have yielded different results. Finally, our tracing of potential pathways from SNAP to purchasing decisions to children’s access and consumption and then to body weight is, by necessity, speculative, in part because we do not have indicators for things in which we are particularly interested, such as where people shop or to what extent children eat at home.

This study focused on neighborhood socioeconomic environment, but, as demonstrated by Cannuscio et al. (2014) and Carroll-Scott et al. (2013), the key neighborhood domains—socioeconomic, built, and social—interact. Because studying built and social environments often demands a level of direct observation that is both time- and resource-intensive, studies of the environments in which SNAP participation, food purchase, and food consumption take place tend to rely on socioeconomic indicators. Those studies that do include information on built environments and social interactions tend to be conducted in limited regions. A major limitation of existing work is that it often assumes people acquire and consume food in the neighborhoods in which they live; future research should consider mobility and the spatial dimensions of food access. We may be reaching a time when, for our understanding of any relationship between SNAP participation and body weight to expand, more holistic, multi-dimensional views of neighborhoods will be needed. Such views have the potential to uncover the mechanisms underlying not only SNAP participation effects, but the effects of individual and collective poverty, social modeling, environmental discrimination, and public policy, thereby allowing governments and communities to better target intervention.

Overall, study results suggest that SNAP is not leading families with children to have food consumption patterns that increase children’s obesity risk. Furthermore, for families in disadvantaged neighborhoods, SNAP may be supporting improved nutritional health and thus lowering obesity risk for children.

Disclosure statement

No potential conflict of interest was reported by the author(s).

Notes on contributors

Thomas P. Vartanian, Ph.D., is a professor at the Graduate School of Social Work and Social Research at Bryn Mawr College. He has published extensively on the SNAP/Food Stamp Programs and on long-term neighborhood effects. He has numerous grants to examine SNAP and TANF, including grants from the Economic Research Service at the USDA, the Institute for Research on Poverty at the University of Wisconsin, the RIDGE Center for Targeted Studies at Purdue University, and the Joint Center for Poverty Research at Northwestern University and the University of Chicago.

Linda Houser, Ph.D., is an associate professor and Ph.D. Program Director at Widener University’s Center for Social Work Education whose practice and research focus on health and health disparities, employment, and caregiving. Recent publications on topics including place-based health disparities, family leave, and work life integration for parents of children with autism have appeared in publications such as Demography, Families in Society, the Journal of Health and Social Behavior, and the Journal of Autism and Developmental Disabilities.