Demography

, Volume 49, Issue 3, pp 1127–1154

The Effects of Childhood SNAP Use and Neighborhood Conditions on Adult Body Mass Index

Authors

    • Graduate School of Social Work and Social ResearchBryn Mawr College
  • Linda Houser
    • Center for Social Work EducationWidener University
Article

DOI: 10.1007/s13524-012-0115-y

Cite this article as:
Vartanian, T.P. & Houser, L. Demography (2012) 49: 1127. doi:10.1007/s13524-012-0115-y

Abstract

The disproportionate number of individuals who are obese or overweight in the low-income U.S. population has raised interest in the influence of neighborhood conditions and public assistance programs on weight and health. Generally, neighborhood effects and program participation effects have been explored in separate studies. We unite these two areas of inquiry, using the 1968–2005 Panel Study of Income Dynamics (PSID) to examine the long-term effects of childhood Supplemental Nutrition Assistance Program (SNAP) participation, neighborhood conditions, and the interaction of these two, on adult body mass index (BMI). Using sibling fixed-effects models to account for selection bias, we find that relative to children in other low-income families, children in SNAP-recipient households have higher average adult BMI values. However, the effects of childhood SNAP usage are sensitive to both residential neighborhood and age at receipt. For those growing up in advantaged neighborhoods, projected adult BMI is higher for children in SNAP-recipient households than for children in low-income, nonrecipient households. In contrast, for those growing up in less-advantaged areas, adult BMI differences between children in SNAP-recipient and those in low-income, nonrecipient households are small. SNAP usage during preschool years (0 to 4) has no impact on adult BMI scores. However, at later childhood ages, the time elapsed receiving SNAP income increases adult BMI values relative to a condition of low-income nonreceipt.

Keywords

NeighborhoodsSNAPFood StampsBMIObesity

Introduction

A period of high U.S. unemployment, beginning in early 2008, has led to a marked increase in enrollment in public assistance programs, although in none so dramatically as in the Supplemental Nutrition Assistance Program (SNAP), formerly the Food Stamp Program (FSP). Between 2007 and 2009, SNAP enrollment increased by over 27%, with benefit costs rising by an estimated 66% (almost $20 billion). Research interest in the effects of participation in this safety net program has extended to numerous economic and nutrition outcomes, including labor supply (Hoynes and Schanzenbach 2007); consumption stabilization (Gundersen and Ziliak 2003); healthy food choices (Frazao et al. 2007); mortality (Krueger et al., 2004); and nutrient availability, dietary quality, and food security (for reviews, see Burstein et al. 2004; and Currie 2003).

Given the dietary and health concerns that spurred expansion of the FSP nationwide in 1974, it is not surprising that two of the most heavily researched participation outcomes have been food security (Gibson-Davis and Foster 2006; Mykerezi and Mills 2008; Yen et al. 2008), and obesity or healthy weight (e.g., Baum 2007). With national attention focused on a recently dubbed “obesity epidemic,” targeting concern particularly to children and low-income families (Drewnowski and Specter 2004; World Health Organization 2010), studies have questioned whether SNAP participation has perversely negative effects on healthy weight. However, as several researchers have pointed out, studies in these areas have been vulnerable to selection bias: that is, to the possibility that families who are eligible and participating and those who are eligible and nonparticipating differ in ways that mask or are mistaken for program participation effects (Yen et al. 2008).

We extend existing research in the area of SNAP participation and weight by using longitudinal data from the Panel Study of Income Dynamics (PSID) to examine the relationship between SNAP participation during childhood and adult body mass index (BMI) in the context of and in interaction with neighborhood conditions. Estimates are derived from the proportion of childhood years spent with SNAP receipt, relative to the proportion of time spent with low income (i.e., annual income under 150% of the poverty line) but without SNAP assistance. We also explore the ways in which selection may bias estimates of SNAP participation effects over the long run, by commenting on differences between sibling fixed-effects estimates and estimates derived from standard regression models. Although previous studies have looked at the relatively short-term effects of SNAP on current participants, this study examines the long-run impacts of SNAP, using longitudinal data and adjusting for selection bias.

Another key contribution of the study is its attention to neighborhood factors. Geographic variation in weight, dietary practice, and food sufficiency is well documented (Robert and Reither 2004). Moreover, the recommendations that may emerge from neighborhood-level findings have arguably the advantage of affecting groups of individuals in ways that are both cost- and time-effective relative to individual-level interventions (Black and Macinko 2008). Thus, we examine whether SNAP participation affects children differently in more- or less-advantaged neighborhoods, paying particular attention to neighborhood affluence, relative disadvantage, and, at the county level, crime. Differences in outcomes between SNAP recipients may be due to differential access to local food sources and neighborhood social and institutional resources, as well as differential exposure to factors, such as crime and violence, that together affect whether SNAP funds are used and, if so, how they are used.

Lastly, in addition to examining the effects of SNAP exposure across childhood years, we examine age brackets that correspond roughly to preschool age (0 to 4), primary school age (5 to 8), early adolescence (9 to 13), and late adolescence (14 to 18). Thus, we address not only whether adult weight is influenced by program participation and neighborhood conditions during childhood, but also at what ages such influence is most pronounced.

Background and Significance

SNAP Participation and BMI

The disproportionate number of individuals who are overweight or obese in the low-income U.S. population has raised interest in the influence of both neighborhood conditions and public assistance programs on weight and health (Drewnowski and Specter 2004). Several competing views attempt to explain differences in long-term weight and health outcomes for SNAP participants relative to low-income nonparticipants, with the majority of such explanations predicting worse outcomes for children from SNAP-recipient families.

One such view suggests that the mechanism through which participants receive their SNAP benefits (i.e., food coupons or “stamps” and, most recently, debit cards) incentivizes overpurchase and thereby overconsumption of food. Relative to cash assistance, SNAP benefits represent an inflexible form of income; the relative price of food is low or non-existent for the value of the SNAP funds received. Using random assignment experiments in which participants are given either cash or SNAP assistance, Fraker et al. (1995) found that food expenditures from food stamps are 18% to 28% higher than are food expenditures from cash assistance. To the extent that the overpurchasing of food leads to overconsumption and thereby to becoming overweight, SNAP assistance may have long-term negative effects on children’s health and healthy weight.

A second view proposes that it is not so much the mechanism of delivery as the periodicity of delivery that may artificially inflate consumption and weight. Ver Ploeg and Ralston (2008) suggested that SNAP participation leads to cyclical periods of abundance (i.e., the beginning of the month, when SNAP funds are received), but also deprivation (i.e., the end of the month, when the allotted benefit has been expended). This model of binge eating, which may become habitual over time, has been shown to lead to long-term weight gains (Polivy et al. 1994).

Conversely, SNAP participation may promote the maintenance of healthy weight among participants by giving them the financial means to choose nutrient-rich and/or low-calorie foods. To the extent that purchase and consumption of healthy foods replaces higher calorie/higher fat but less-expensive foods, this perspective would predict lower BMI values for SNAP recipient households relative to low-income, nonrecipient households. However, Wilde and Ranney (2000) failed to find significant differences between the diets of SNAP participants and low-income nonparticipants. Moreover, as explored further herein, the extent to which SNAP availability affects food choices may be heavily conditioned by geographic variation in the availability and cost of various foods.

To date, SNAP research has reported mixed results for the association between program participation and BMI values. Most cross-sectional studies do not find an association between SNAP use and obesity among children, regardless of gender (Bhattacharya and Currie 2000; Boumtje et al. 2005; Ver Ploeg et al. 2007). When analyzing participation outcomes for adults, however, Townsend et al. (2001) reported a positive relationship between SNAP participation and obesity for adult women aged 20 and older but not for men, while Ver Ploeg et al. (2007) found little or no relationship for either women or men.

Using longitudinal data, Jones and Frongillo (2006) and Kim and Frongillo (2007) found no relationship between SNAP participation and being overweight. However, when longitudinal data are paired with efforts to address selection bias, employing either person-specific fixed-effects or instrumental variables approaches, researchers tend to find more evidence for a relationship between SNAP participation and being overweight, especially among women (Baum 2007; Chen et al. 2005; Gibson 2003; Meyerhoefer and Pylypchuk 2008). Gibson (2004), using individual fixed-effects models, found strong, positive program participation effects on being overweight for girls, especially for those girls who use SNAP assistance for an extended time, but found negative effects for boys. For men, Baum (2007) found that fixed-effects models reveal positive relationships between SNAP participation and BMI; Gibson (2003), Meyerhoefer and Pylypchuk (2008), and Chen et al. (2005) detected no such effects for men.

Most of these studies have used either cross-sectional data or longitudinal data that cover a relatively short period of time—generally, a small number of years during childhood, adolescence, or early adulthood. None have examined the long-term effects of childhood SNAP participation into adulthood. In addition, none of the aforementioned studies considered differential effects by neighborhood.

Neighborhoods and BMI

Our interest lies in exploring not only the long-term effects of childhood SNAP participation but also how these effects may be conditioned by area of residence. Although debate continues over the mechanisms by which neighborhoods affect various outcomes, and thus over the measures best suited to capture meaningful effects, substantial research evidence exists that residential neighborhoods are associated with both specific health outcomes (McGrath et al. 2006) and health-related conditions, including BMI scores (Burdette and Whitaker 2005) and obesity (Black and Macinko 2008). We wish to emphasize, though, that our concern here is not only with the ways that attributes such as neighborhood safety or inadequate funds for quality play spaces may limit opportunities for outdoor play, thereby increasing weight (Burdette and Whitaker 2005); we also examine the ways that neighborhood factors may limit opportunities for using SNAP benefits, with implications for maintaining healthy weight (Moore and Diez Roux 2006).

Neighborhood Affluence and Relative Disadvantage

Much of the neighborhood effects literature assumes that, when all other relevant factors are controlled, residing in high-income or otherwise advantaged areas will yield better outcomes than will residing in low-income or otherwise disadvantaged areas (Jencks and Mayer 1990). Neighborhoods with high levels of affluence and socioeconomic cohesion have the capacity to attract and sustain high-quality and reliable basic institutions—resources that tend to elude poorer communities (Wilson 1987). Thus, healthy weight may be linked to neighborhood-level affluence through access to better food sources (e.g., stores that boast lower costs and fresher foods), to safer places to play, and to models for healthy eating and nutrition education (e.g., neighbors, community wellness programs) (Moore and Diez Roux 2006; Powell et al. 2007).

In recent years, the term “food desert” has emerged to designate an area in which residents lack access to affordable supplies of those foods considered essential for a healthy diet (Beaulac et al. 2009). In a 2010 review of 132 studies on food access and health, PolicyLink and the Food Trust together concluded that evidence in support of the existence of U.S. “food deserts” is overwhelming: communities that are predominantly of color, low-income, or rural face a disproportionate degree of difficulty accessing healthy foods (Treuhaft and Karpyn 2010). Further compounding concerns with food scarcity, such communities often have outdated, unreliable, or inaccessible systems of transportation, effectively compounding both the time and monetary costs of shopping outside the neighborhood. In a particularly telling study, Powell et al. (2007) reported that low-income neighborhoods have 25% fewer chain supermarkets but 1.3 times as many convenience stores relative to middle-income areas. According to Macintyre (2007), deprivation under such conditions may be “amplified”: low-income families who are already “priced out” of the purchase of healthy foods are further disadvantaged by structural limits on access.

There is noteworthy evidence, too, that “access” is linked both to “choice” and to BMI. Among low-income families, the relatively high cost of healthy foods has been cited as a barrier to choosing a healthy diet (Hamelin et al. 2002; Jetter and Cassady 2006). Of particular relevance to our study of impacts accrued during childhood, Lytle et al. (1996) reported that comprehensive changes in food service and classroom education lead to healthier school lunch intake in children. For elementary school children, and for children in poverty and children of color in particular, lower prices for fruits and vegetables have been linked to lower BMI scores (Sturm and Datar 2005).

Taken together, these studies suggest that neighborhoods may be placed on a continuum from relative affluence to relative poverty, with optimal health and weight outcomes found on the side of affluence. This does not mean, however, that health and weight outcomes for low-income families residing in relatively advantaged neighborhoods will be uniformly positive in a qualitative rather than a statistical sense. Living in more-advantaged neighborhoods may give low-income families greater food access generally and at lower cost overall than would be found in less-advantaged areas (e.g., in “food deserts”). Having limited funds to spend on food may lead to a greater consumption of high-calorie, high-fat foods in areas with high food access. Moreover, some studies suggest that even after access to affordable, healthy foods is expanded in “food deserts” (e.g., farmers’ markets in urban neighborhoods), dietary practices show little change; previously established consumer preferences and habits persist (Cummins et al. 2005).

Similarly, several recent health studies have questioned whether living in an affluent area is uniformly “good” for resident health, particularly when individual families find themselves socioeconomically disadvantaged relative to their neighbors (Pham-Kanter 2009; Vartanian and Houser 2010). Those who are poor relative to some neighborhood “norm” may have less favorable outcomes because of psychosocial strain created by persistent reminders of their “relatively disadvantaged status;” difficulty competing or even socializing within a relatively advantaged context (Galea and Ahern 2005); or an absence of community resources structured to serve and accustomed to serving low-income residents.

SNAP receipt may further distinguish low-income recipient children from their nonrecipient neighbors with ramifications for adult outcomes. Such a differentiation could take several forms. SNAP-recipient children may experience stigma and shame leading to isolation and poor health or weight outcomes (Sobal 1991). Institutional factors may also play a role if locally prevalent food sources are unaccustomed to accepting or are ill-equipped to accept SNAP debit cards, or if greater access to even healthy foods increases consumption to the point of overweight.

Social Disorganization

In addition to neighborhood affluence and “relative disadvantage” literature, our analysis draws from Shaw’s and McKay’s (1942) description of social disorganization theory as a framework for understanding crime—and, more recently, for predicting health outcomes (Cagney et al. 2005). Shaw and McKay identified multiple, interacting mechanisms of community-level disadvantage, including poverty and racial isolation, thought to work together to constrain access to resources and interfere with the formation of communal bonds. Of particular relevance to our study and to the construction of a comprehensive measure of neighborhood advantage, racial segregation has been associated with both higher BMI scores and increased odds of being overweight (Chang 2006).

Social disorganization theory also supports the inclusion of measures of community crime or violence in models predicting health and weight outcomes. Lumeng et al. (2006) found a negative association between parental perceptions of neighborhood safety and obesity in school-age children. Area-level crime, generally seen as operating through heightened levels of stress and limited opportunities to leave the relative safety of home, reduces opportunities for physical activity and heightens susceptibility to adverse health outcomes (Black and Macinko 2008). There is also evidence that chronic stress affects weight via stress-induced eating and a preference for foods high in sugar and fat (Torres and Nowson 2007).

As we describe further, the community crime measure that we use has limitations; thus, this aspect of our analysis reflects a limited operationalization of a complex and encompassing theory.

Age Effects

The success of efforts to translate research evidence into policy or practice recommendations for promoting healthy weight depends in part on deepening our understanding of which factors are most likely to function in protective ways, as well as when, in the life course of children, effects are most likely to accrue. Some potential explanations for a childhood program participation effect depend on the child’s awareness of family participation; others require only that the child’s health or weight be influenced by family’s SNAP use. This distinction alone suggests that age is an important dimension of study.

In general, parent or caregiver influence is thought to dominate during children’s early years, giving partial way to the influences of peers, schools, and neighborhoods as children enter grade school and beyond (Erikson 1997). As a result, researchers exploring the ways in which neighborhood conditions during childhood affect future health outcomes tend to focus primarily or solely on the presumed vulnerability of adolescents (McGrath et al. 2006; Plotnick and Hoffman 1999). Focusing on the adolescent years tends to be justified as well by a presumption that neighborhood effects are either cumulative or lagged and, thus, difficult to detect prior to adolescence (Wheaton and Clarke 2003).

As Wheaton and Clarke discovered, however, in their study of early adult mental health, neighborhoods may exert effects on parents that show up in their children at even very young ages. This is certainly plausible in the area of nutrition and weight, as children at young ages are almost entirely dependent on parents’ food choices and, thus, wholly susceptible to neighborhood-based limitations. In fact, it could be argued that children at school age and beyond should be less susceptible to neighborhood effects on nutrition and weight, as they begin to be fed “outside their neighborhood” (e.g., at school, where in theory, standards of nutritional adequacy prevail).

To examine children’s sensitivity to program participation and neighborhood effects at various ages, we adopt an approach used previously by Duncan et al. (1998), Levy and Duncan (2000), and Vartanian and Buck (2005). Specifically, we examine age brackets that correspond roughly to preschool age (0 to 4), primary school age (5 to 8), early adolescence (9 to 13), and late adolescence (14 to 18). These studies found that their respective variables of interest exert strong and statistically significant effects on adult outcomes, especially when measured at the earliest ages (i.e., 0 to 4).

Data and Methods

We draw data from the 1968–2005 Panel Study of Income Dynamics (PSID), a longitudinal data set that, as of 2005, includes over 8,000 families and almost 23,000 individuals. The PSID was collected yearly from 1968 to 1997, then biannually for years 1999–2005. By merging four decades of U.S. census data with the PSID, we incorporate information on PSID respondents’ neighborhoods, herein defined as their census tracts. We merge 1970 census information onto PSID years 1968–1975, 1980 census information onto PSID years 1976–1985, 1990 census information onto PSID years 1986–1995, and 2000 census information onto PSID years 1997–2005.

To be included in the sample, individuals must have a minimum of four years of childhood data (age 0 to 18), and between 1 and 34 years of adult data. Kunz et al. (2003) concluded that childhood neighborhood characteristics are so highly correlated across childhood years that having even a single year of neighborhood information produces only small errors-in-variables bias. The maximum respondent age is 52 (i.e., age 15 in 1968, with participation through the 2005 PSID wave).

Dependent Variables

Our dependent variable is adult BMI. The PSID began asking both heads of households (either male or female) and wives their weight and height in 1999. From this, we derived average BMI values from 1999 until 2005, or whenever the individual left the sample.

The fact that height and weight are self-reported introduces the potential for measurement error. Merrill and Richardson (2009) found that self-reported height tends to be overstated for both men and women (especially at older ages), and weight tends to be understated (especially by women). Thus, the results reported here likely understate BMI values. There is also evidence, however, that self-reported height and weight are adequately reliable indicators of clinical measures (Voaklander et al. 2006). Adams et al. (2006) reported that self-assessed height and weight together predict mortality, although perhaps more weakly in older adults than in younger adults. Mehta and Chang (2009) similarly detected understating of self-reported BMI values using National Health and Nutrition Survey (NHANES) data, which also include a clinically measured BMI value. Nonetheless, they argued that these underestimates do not affect their overall conclusions that BMI values above 35 increase mortality for both middle-aged males and females by 62% and 40%, respectively. In another analysis of NHANES data, Gorber and Tremblay (2010) concluded that the discrepancy between self-reported and clinically measured BMI is modest at 3%, and has been stable during the past 30 years.

Independent Variables

Government Program Participation (SNAP and AFDC)/(TANF)

Capitalizing on the longitudinal advantages of the PSID sample, we conceptualize program participation as exposure across childhood years and across adult years. We examine the percentage of childhood years (a) using SNAP assistance alone; (b) using SNAP and TANF assistance jointly; (c) having income at or below 150% of the poverty line but not receiving either TANF or SNAP assistance (hereafter, “low-income nonreceipt”); and (d) having income above 150% of the poverty line (hereafter, “non-eligible nonreceipt”). Regression models exclude the proportion of time with low-income nonreceipt.

Childhood and Adult Neighborhood

We use three primary indicators of neighborhood quality in childhood and adulthood: a neighborhood advantage index, a measure of relative deprivation, and county-level crime rate.

The neighborhood advantage index is constructed using principal components analysis with six neighborhood variables, including (a) poverty rate, (b) percentage of households with incomes above $35,000 and (c) above $60,000, (d) percentage of female-headed households, and (e) percentage of white and (f) of black residents. Each component indicator is highly correlated with the principal component variable, with absolute values of the correlation above .79 for all observations; the neighborhood advantage index explains over 76% of the variance of its component variables. Similar results were obtained in the construction of the adult neighborhood advantage index. Higher index values signal more-advantaged neighborhoods.

The relative deprivation measure consists of the percentage of neighborhood residents in income brackets above that of the focal individual, relative to the percentage of neighborhood residents in the same or lower bracket.

In the absence of census tract or neighborhood-level data on crime or violence, the final indicator—crime rate—is measured at the county level, using data collected by the U.S. Department of Justice, FBI, Uniform Crime Reporting Program, and published in County and City Data Books (CCDB). These data are available for the years between 1967 and 2007. We merge county crime data from the closest available year to each PSID year (e.g., data for the 1968 to 1970 years of the PSID are merged with crime data from the 1967 Data Book). Because the crime data are recorded at a higher level of aggregation (i.e., county rather than census area), and because this is a first effort at using county-level crime data to examine SNAP use and weight outcomes, we do not include crime statistics in the neighborhood advantage index.

We interact neighborhood and program participation variables to determine whether childhood SNAP usage affects BMI levels differently by the type of neighborhood where one grows up. First, we examine the interaction between the average amount of time receiving SNAP income and the average neighborhood condition. Second, we examine the interaction of SNAP usage by the neighborhood condition for each childhood year, and average these over the childhood years. Although we use two different methods to examine these interactions, we report only on the first method because the two methods yield highly similar results.

Covariates used in each of the models are listed in Table 1.
Table 1

Weighted means and standard deviations

 

Sibling Sample

Full Sample

Mean

SD

Mean

SD

Adult Weight Outcomes

 BMI

28.53

6.48

28.45

6.48

 Normal weight

0.29

0.46

0.30

0.46

 Overweight

0.34

0.48

0.34

0.47

 Obese

0.34

0.47

0.34

0.47

 BMI women

28.29

6.95

28.20

6.82

 BMI men

28.79

6.64

28.71

6.80

Childhood Neighborhood

 Neighborhood index

0.00

1.00

0.00

1.00

 Proportion in poverty

0.12

0.10

0.13

0.10

 Proportion with income > $35 K

0.73

0.15

0.72

0.15

 Proportion black

0.11

0.21

0.11

0.21

 Proportion with income greater than respondent’s

0.37

0.42

0.55

0.56

 Crime rate

2.21

2.26

2.18

2.46

Adult Neighborhood

 Neighborhood index

0.00

1.00

0.00

1.00

 Proportion in poverty

0.12

0.08

0.13

0.08

 Proportion with income > $35 K

0.70

0.15

0.70

0.15

 Proportion black

0.12

0.18

0.13

0.19

 Crime rate

0.60

0.84

0.71

1.02

Program Participation as % of Childhood Time

 % of time with SNAP income

0.06

0.13

0.06

0.13

 % of time with SNAP and TANF

0.06

0.18

0.06

0.18

 % of time with low-income nonreceipt

0.15

0.23

0.15

0.23

 % of time with high-income nonreceipt

0.73

0.36

0.74

0.35

Childhood Covariates

 Health index, child

0.93

0.15

0.92

0.15

 White

0.83

0.37

0.82

0.38

 Black

0.13

0.34

0.14

0.35

 Other race

0.04

0.18

0.04

0.19

 Sex (female)

0.52

0.50

0.52

0.50

 Number of children in family

2.94

1.34

2.65

1.34

 Age of youngest child

7.42

2.53

7.49

2.57

 Birth order

2.12

1.19

1.86

1.13

 Age of household head

40.62

0.75

40.70

8.00

 Head: High school dropout

0.22

0.38

0.23

0.42

 Head: High school graduate

0.18

0.37

0.18

0.39

 Head: Some college

0.15

0.36

0.15

0.35

 Head: College graduate

0.44

0.50

0.44

0.50

 Proportion of years with a move

0.15

0.18

0.15

0.18

 Family income-to-needs

2.74

1.83

2.80

1.84

 Income variance

1.38

12.12

1.27

10.83

 Head: Any health limits to work

0.07

0.26

0.08

0.27

 Value of house ($000,000)

1.33

1.00

1.32

1.09

 % Years owning a house

0.53

0.30

0.51

0.31

 % Time married

0.86

0.28

0.84

0.28

 % Time never married

0.01

0.09

0.02

0.11

 % Time divorced

0.07

0.19

0.08

0.20

 % Time separated

0.03

0.12

0.03

0.12

 % Time widowed

0.03

0.14

0.03

0.14

 Residence: Big city

0.24

0.43

0.24

0.43

 Residence: Urban

0.24

0.43

0.24

0.43

 Residence: City

0.13

0.34

0.13

0.34

 Residence: Suburban

0.10

0.30

0.10

0.30

 Residence: Rural

0.10

0.30

0.10

0.30

 Residence: Very rural

0.19

0.39

0.19

0.39

 Region: South

0.27

0.44

0.28

0.45

 Number of childhood years in sample

12.41

4.49

12.37

4.64

 Age at the end of the sample

38.84

8.15

38.19

8.84

 Person became head/wife before age 18

0.02

0.15

0.02

0.15

 Ever smoked

0.37

0.48

0.38

0.48

 Focal individual: High school dropout

0.12

0.32

0.11

0.31

 Focal individual: High school graduate

0.25

0.43

0.25

0.43

 Focal individual: Some college

0.21

0.40

0.22

0.41

 Focal individual: College graduate

0.43

0.50

0.43

0.49

 Year started: 1968–1972

0.72

0.44

0.68

0.47

Adult Covariates

 % of sample with SNAP income

23.28

46.47

24.04

46.60

 Average SNAP income

121.15

414.05

117.97

397.17

 Family size

2.70

1.06

2.67

1.07

 Family income-to-needs

3.40

2.29

3.33

2.30

 Marital status: Married

0.57

0.37

0.56

0.37

 Marital status: Never married

0.32

0.38

0.34

0.39

 Marital status: Widowed

0.00

0.03

0.00

0.04

 Marital status: Divorced

0.08

0.17

0.07

0.16

 Marital status: Separated

0.03

0.09

0.03

0.09

 Live in an metropolitan area

0.53

0.50

0.54

0.50

Statistics for SNAP-Only Use (variables not used in regression models)

 % Ever using SNAP in childhood

0.28

0.45

0.27

0.44

 % Ever using SNAP in adulthood

0.23

0.42

0.24

0.43

 % Using SNAP only in childhood

0.16

0.36

0.15

0.36

 % Using SNAP only in adulthood

0.12

0.32

0.12

0.33

 % Using in SNAP in both adulthood and childhood

0.12

0.32

0.12

0.32

 % Not using SNAP in either childhood or adulthood

0.61

0.49

0.61

0.48

N

 

3,306

 

4,658

Construction of Child and Adult Predictors

For all variables, we calculate the average value of the characteristic during the childhood or adult years, respectively. The first set of models includes only childhood variables; the second set incorporates variables from the focal respondents’ adult years. For models that examine age ranges during childhood, we average variable values over only those ages.

For models that use all observations, the sample size is 4,658. For models restricted to sibling groups, the sample size is 3,306 observations.

Analytic Strategy

As noted earlier, studies of program participation may be confounded by self-selection into the SNAP program. For example, there may be shared family norms surrounding the use of government assistance that are not immediately manifest in reports of current use, yet that exist in unreported form nonetheless (Linz et al. 2005). To control for shared attributes of siblings, both biological and environmental, we use sibling fixed-effects models (Dietz 2002; Leventhal and Brooks-Gunn 2000).

Three limitations attached to the use of fixed-effects models are important to note. First, models are limited to individuals with siblings, excluding those from single-child households or from households in which only one child has valid variable values. Second, and related to this first point, sample sizes are likely to be smaller than with standard regression models; therefore, estimates may be less precise. Third, unobserved factors that vary over children are not captured. We respond to the first two of these limitations by running standard regression ordinary least squares (OLS) models, first using all observations and then only those observations with siblings. If estimates from sibling and nonsibling OLS models are similar, we can have greater confidence that the fixed-effects results are not being driven primarily by the sibling sample restriction.

Because the effects of SNAP participation may differ substantially by gender (Ver Ploeg and Ralston 2008), we also comment on gender-specific estimates.

Results

Descriptive Findings

According to Table 1, few descriptive differences exist between the full sample and the subsample of siblings. For example, average adult BMI is 28.53 for the full sample and 28.45 for the sibling sample. Going forward, we discuss only the sibling sample, the source of our main regression results.

In adulthood, 34% of respondents are overweight (BMI: 25–29.9), with roughly the same percentage who are obese (BMI: 30 and above), while 29% are of normal weight (BMI: 18.5–24.9). These values roughly mirror the 2003–2004 NHANES U.S. prevalence estimates of 34.1% overweight, and 37.0% obese or extremely obese (Ogden and Carroll 2010). Men have slightly higher average BMI values than women.

Because we are working to disentangle SNAP effects from TANF effects, the SNAP data we report below are for SNAP-only use. The full sample average for childhood time receiving SNAP-only income is roughly 6%. However, among the 28% of respondents who receive SNAP income at some point during childhood, the average amount of childhood time with receipt is 25%. Twenty-three percent of respondents receive some SNAP income as adults, averaging $617 in aid per year (results not shown). The overlap between childhood recipients and adult recipients is substantial but not total: 12% of the sample reports receiving SNAP in both childhood and adulthood, with 16% receiving only in childhood and 12% receiving only in adulthood.

Sibling fixed-effects models rely on differences between siblings on variables of interest to estimate effects. To contextualize the sibling models for this study, Table 2 details differences in sibling values for key study variables, both for the full sibling sample and for subsamples of those whose families spend some time in a particular condition (e.g., receiving SNAP income). Overall, over 41% of siblings have different values for the percentage of childhood spent receiving SNAP; most of those without differences share a value of zero (i.e., their families received no SNAP income during their childhood years). For those families with some SNAP income, 96.18% of siblings have different values for the percentage of time with SNAP receipt, with an average difference of 5.53%. Our use of childhood exposure to SNAP, rather than a simple distinction between recipient and nonrecipient families, allows us to capitalize on differences between siblings that accumulate throughout their respective childhood years. Although we cannot entirely rule out measurement error as a source of between-sibling variation, the likelihood of recall bias is reduced by the fact that childhood variable values are based, not on retrospective reports from adults, but instead on reports from parents during the year in question. Between-sibling differences may result from changing family circumstances (e.g., moving from SNAP receipt as a young family to no receipt later in the family’s life cycle), changing neighborhood conditions, and family residential moves. For each of the three primary neighborhood indicators, over 96% of siblings have different values. Nearly all siblings (i.e., 99.5%) have different adult BMI values.
Table 2

Between-sibling differences on variables of interest

 

% of Sibling Groups With Nonzero Differences

% With Nonzero Differences for Families in the Condition

% Within-Family Difference for Those With Nonzero Differences

M

SD

Childhood

 % Time with SNAP

41.11

96.18

5.53

5.85

 % Time with SNAP/TANF

23.29

92.21

1.83

5.20

 % Time with low-income nonreceipt

59.59

95.44

6.50

6.90

 % Time with high-income nonreceipt

55.02

63.25

3.60

6.07

 Neighborhood index

96.55

––

0.15

0.22

 Crime rate/1,000

96.58

––

2.20

2.53

 Proportion with income greater than respondent’s

99.03

––

3.08

3.45

Adulthood

 BMI

99.50

––

3.72

3.30

 Neighborhood index

97.28

––

0.63

0.62

Note: Sample size = 3,306.

Regression Results

Standard Regression Models

We run standard regression models (OLS with robust standard errors) first with all observations and then with siblings only to determine whether restricting the sample to siblings may pose problems for the fixed-effects analyses. Although we show results for only the neighborhood index on Table 3, we found similar results for all other model specifications. Namely, although the model using all observations suggests that childhood SNAP receipt has a statistically significant positive effect on adult BMI (+2.67; p < .01), the siblings-only model understates this effect (+2.16; p < .10). Moreover, although the all-observations model indicates that childhood neighborhoods compound the SNAP effect, the neighborhood-by-SNAP receipt interaction coefficient in the siblings-only model, although similar in both size and direction, does not reach the level of statistical significance. Together, this cautions early on that restricting the sample to siblings in fixed-effects models may underestimate the primary effects of interest. However, it is important to note that the effect is not distorted; both the size and direction of the coefficients for the two models are consistent.
Table 3

Childhood-only OLS results with robust standard errors, using the neighborhood index

 

All Observations

Siblings Only

B

SE

B

SE

Neighborhood Index

−0.21

(0.19)

−0.11

(0.24)

Neighborhood Index × % Time With SNAP

1.76

(0.72)**

1.09

(0.85)

% Time With SNAP

2.67

(0.99)**

2.16

(1.19)

% Time With SNAP and TANF

−0.25

(0.65)

−0.82

(0.79)

% Time With High-Income Nonreceipt

0.20

(0.54)

−0.03

(0.64)

N

 

4,658

 

3,306

Note: Higher neighborhood index values indicate more advantaged neighborhoods. Each model also includes all childhood-level covariates as shown in Table 1.

p < .10; **p < .01 (two-tailed tests)

Fixed-Effects Models Using Childhood Covariates Alone

When we control for unobserved but nonvarying factors among siblings as well as a full set of covariates, we find that individuals who grow up in households with SNAP usage across childhood years are predicted to have adult BMI values that are 9.38 units higher than those of individuals growing up in low-income households with no SNAP usage (Table 4). Further, although childhood neighborhood of residence (i.e., neighborhood advantage index) has no independent effect on adult BMI, the interaction between neighborhood advantage and SNAP program participation is statistically significant, suggesting that growing up in advantaged neighborhoods adds to the positive effect of childhood SNAP usage on adult BMI.
Table 4

Sibling fixed-effects models

 

Neighborhood Index

% With Incomes Above the Respondent’s

Crime Rate

B

SE

B

SE

B

SE

B

SE

B

SE

B

SE

Childhood Variable–Only Models

 Neighborhood variable

−0.57

(0.99)

−0.27

(1.05)

−0.04

(0.75)

0.33

(0.79)

−0.88

(0.59)

−0.87

(0.61)

 Neighborhood variable × % Time with SNAP

5.04

(2.34)*

5.74

(2.82)*

1.04

(2.36)

−0.71

(2.86)

1.14

(1.70)

−0.41

(1.89)

 % Time with SNAP

9.38

(3.05)**

10.75

(3.44)**

4.58

(3.08)

7.17

(3.75)

5.27

(2.43)*

5.63

(2.76)*

 % Time with SNAP and TANF

2.56

(2.37)

1.68

(2.51)

3.14

(2.36)

2.09

(2.57)

3.47

(2.37)

2.43

(2.57)

 % Time high-income nonreceipt

1.22

(1.98)

1.49

(2.12)

0.68

(2.13)

1.80

(2.23)

0.68

(1.96)

1.27

(2.06)

 Female × % Time with SNAP × Neighborhood variable

  

−1.46

(2.15)

  

2.61

(2.36)

  

3.41

(2.04)

 Female × Neighborhood variable

  

−0.30

(0.47)

  

−0.67

(0.45)

  

0.22

(0.33)

 Female × % Time with SNAP

  

−2.71

(2.81)

  

−4.10

(3.13)

  

−0.69

(2.33)

 Female × % Time with SNAP and TANF

  

1.27

(1.71)

  

1.03

(1.55)

  

0.62

(1.54)

 Female × % Time high-income nonreceipt

  

−0.72

(1.23)

  

−2.30

(1.29)

  

−1.12

(1.11)

Child and Adult Variable Models

 Neighborhood variable

−0.56

(1.04)

−1.16

(1.12)

−0.15

(0.78)

0.06

(0.82)

−0.86

(0.59)

−0.78

(0.62)

 Neighborhood variable × % Time with SNAP

5.34

(2.46)*

6.89

(2.95)*

1.63

(2.48)

−0.71

(3.07)

1.75

(1.71)

−0.75

(1.94)

 % Time with SNAP

9.89

(3.16)**

11.87

(3.60)***

4.61

(3.16)

7.65

(3.86)*

5.19

(2.45)*

6.26

(2.88)*

 % Time with SNAP and TANF

1.24

(2.45)

0.56

(2.65)

2.06

(2.44)

1.14

(2.65)

2.87

(2.37)

1.33

(2.64)

 % Time neither high-income nonreceipt

1.00

(2.09)

0.91

(2.23)

0.62

(2.22)

0.08

(2.34)

0.60

(1.97)

0.02

(2.19)

 SNAP as an adult ($1,000)

1.07

(0.52)*

2.83

(0.91)**

1.54

(0.51)**

2.84

(0.91)**

1.32

(0.43)**

2.72

(0.91)**

 Neighborhood variable, adult

−0.42

(0.30)

0.03

(0.41)

0.13

(0.25)

−0.08

(0.39)

0.33

(0.23)

−0.16

(0.39)

 SNAP × Neighborhood variable, adult

0.10

(0.13)

0.45

(0.31)

−0.72

(0.41)

0.46

(0.31)

0.78

(0.73)

0.43

(0.31)

 Female × SNAP ($1,000), adult

  

−1.85

(1.04)

  

−1.88

(1.04)

  

−1.72

(1.04)

 Female × Neighborhood variable, adult

  

−0.55

(0.24)*

  

−0.43

(0.20)*

  

−0.38

(0.20)

 Female × SNAP, adult × Neighborhood variable, adult

  

−0.37

(0.34)

  

−0.40

(0.34)

  

−0.35

(0.34)

Notes: Higher neighborhood index values indicate more-advantaged neighborhoods. For the childhood variable-only results, all childhood covariates were included in the regression models. For the child and adult variable models, all childhood and adult covariates were included in the regression models. N = 3,306.

p < .10; *p < .05; **p < .01; ***p < .001 (two-tailed tests)

We find no evidence that either relative disadvantage (i.e., percentage of residents with income above the respondent’s) or the county crime rate have independent or interactive effects on adult BMI. However, in models using these neighborhood variables, rather than the neighborhood index, the direct effects of SNAP receipt are somewhat weaker, although still positive and statistically significant (see Table 4, columns 2 and 3).

Looking now to the set of models testing interactive effects by sex, we find coefficient estimates that are almost exclusively negative and relatively large, although not statistically significant. The direction of these coefficients suggests that adult weight for females may be less affected by childhood SNAP usage than is adult weight for males. As further support for findings using the neighborhood advantage index, we replace the index with each of the neighborhood indicators that comprise it in turn, and find results that mirror those reported for the index (results not shown).

Fixed-Effects Models Using Childhood and Adult Covariates

The bottom portion of Table 4 depicts results obtained with the addition of adult covariates. Few differences exist between these and the childhood variable-only models. The childhood and adult models do suggest that adult SNAP income has a positive effect on BMI; for each $1,000 increase in SNAP income, BMI is predicted to increase by 1.07 points. What is perhaps most important for our analysis, however, is the robustness of the childhood SNAP effect even with the addition of adult SNAP income. We find support for the conclusion that SNAP usage during childhood has a positive effect on adult BMI, net any adult SNAP effect.

Interactive fixed-effects models using covariates from both childhood and adulthood suggest that women are less affected by SNAP usage in adulthood than are men. For each $1,000 of SNAP income in adulthood, men’s BMI values are predicted to increase by 2.83 points; these adult estimates differ little by which neighborhood variables are used. For women, the increase in BMI per $1,000 of SNAP income is smaller, at slightly less than 1 point.

Table 5 and Fig. 1 show the predicted values of adult BMI for those growing up in neighborhoods that are 1 standard deviation above or below the mean for the neighborhood advantage index. Predictions are derived holding all variables constant except the set of variables for SNAP receipt, sex, and neighborhood conditions (including all interactions). We use three different assumptions to derive these estimates. In the first scenario, we assume that the individual grew up in a low-income household but has not received any SNAP income either as a child or as an adult (column 1). In the second scenario, we hold both childhood and adult SNAP receipt at the mean value for those who receive SNAP income. Specifically, the individual grows up in a low-income household and receives SNAP income for 25% of her childhood, but not for the other 75% of her childhood. Then, during adulthood, she receives $617 in SNAP income, which is the mean level of SNAP income for adults who receive SNAP assistance (column 2). The third scenario holds both childhood and adult SNAP receipt at 1 standard deviation above their mean for those who receive any childhood or adult SNAP income. Specifically, the individual receives SNAP funds for 43% of his childhood (spending 57% of his time with low income but no SNAP), and receives $1,443 in SNAP income per year in adulthood (column 3).
Table 5

Weighted BMI predictions, using the neighborhood index

 

Assumes Low Income but No SNAP Use as a Child or Adult

Assumes 25% Use of SNAP as a Child, and $617/Yearly as an Adult

Assumes 43% Use of SNAP as a Child, and $1,443/Yearly as an Adult

Childhood-Only Variables

 +1 SD, all

25.78

29.38

33.54

 −1 SD, all

26.92

28.00

30.63

 +1 SD, girls

26.03

29.13

32.44

 −1 SD, girls

27.22

28.15

29.80

 +1 SD, boys

25.89

30.06

37.73

 −1 SD, boys

26.46

27.72

32.52

Child and Adult Variables

 +1 SD, all

25.31

29.84

30.80

 −1 SD, all

27.27

29.01

29.81

 +1 SD, girls

27.87

29.54

29.15

 −1 SD, girls

27.28

28.62

28.92

 +1 SD, boys

25.46

31.91

34.43

 −1 SD, boys

27.26

29.82

31.63

Notes: Predictions are derived from the regression models presented in Table 4. For the neighborhood index, higher values indicate more-advantaged neighborhoods. For some models, the interaction between female and the main independent variable is not statistically significant, although some coefficients are relatively large.

https://static-content.springer.com/image/art%3A10.1007%2Fs13524-012-0115-y/MediaObjects/13524_2012_115_Fig1_HTML.gif
Fig. 1

Predicted adult BMI under various assumptions about SNAP receipt and neighborhood of residence, all observations (Table 5)

Moving from a condition of low income but no SNAP receipt, to conditions of more extensive receipt during childhood, we find that predicted adult BMI increases. Children who grow up in low-income households in neighborhoods that are 1 standard deviation above the mean for neighborhood advantage have predicted adult BMI scores of 25.78, slightly above the level of overweight. By contrast, children who grow up in low-income households in neighborhoods 1 standard deviation below the mean for neighborhood advantage have predicted adult BMI scores of 26.92. As our assumptions about SNAP usage change, however, so do our adult BMI predictions. Overall, as we increase the proportion of childhood time spent with SNAP, we find that those growing up in more-advantaged neighborhoods experience increases in predicted adult BMI values that are substantially higher than for those in less-advantaged areas (to 33.54 in above-average childhood neighborhoods, and to 30.63 in below-average childhood neighborhoods, both of which are above the level denoting obesity; see shaded region of Fig. 1).

Using both child and adult SNAP usage in predictions, we again find that those with SNAP income, both in childhood and adulthood, have higher adult BMI values. However, much of this difference is driven by those in more-advantaged neighborhoods. Those growing up in less-advantaged areas experience more modest increases in BMI with increased use of SNAP (going from 27.27 to 29.01 to 29.81, in scenarios assuming increasing degrees of receipt). We also find that adult BMI scores for men are more sensitive to childhood and adult SNAP usage than they are for women (see Fig. 2).
https://static-content.springer.com/image/art%3A10.1007%2Fs13524-012-0115-y/MediaObjects/13524_2012_115_Fig2_HTML.gif
Fig. 2

Predicted adult BMI under various assumptions about SNAP receipt and neighborhood of residence, observations by sex (Table 5)

Regression Effects by Age

To investigate whether childhood effects are being driven by susceptibility at a particular age or ages, we present a series of sibling fixed-effects models for age brackets that correspond roughly to preschool age (0 to 4), primary school age (5 to 8), early adolescence (9 to 13), and late adolescence (14 to 18) (Table 6). Our results suggest that SNAP usage has no statistically discernable effect on adult BMI until the child is older than 4, with the positive effect increasing dramatically at primary school age, and then diminishing somewhat in both early and late adolescence. As indicated earlier, these positive effects on BMI continue into adulthood.
Table 6

Sibling fixed-effects models by age group, using neighborhood index

 

Ages 0–4

Ages 5–8

Ages 9–13

Ages 14–18

B

SE

B

SE

B

SE

B

SE

Childhood-Only Variables

 Neighborhood index

−0.55

(0.73)

−0.70

(0.72)

−0.16

(0.63)

−0.63

(0.57)

 Neighborhood index × % Time with SNAP

−0.74

(2.86)

10.13

(2.79)***

3.72

(2.33)

3.10

(2.02)

 % Time with SNAP

1.58

(5.25)

12.06

(4.63)**

9.10

(3.27)**

7.76

(2.90)**

 % Time with SNAP and TANF

5.49

(4.86)

4.18

(4.18)

3.83

(2.77)

1.04

(2.37)

 % Time with high-income nonreceipt

−3.00

(3.54)

3.78

(3.21)

1.44

(2.28)

2.16

(1.99)

With Female Interactions

 Neighborhood index

−0.63

(0.81)

−0.42

(0.80)

−0.24

(0.70)

−0.41

(0.65)

 Neighborhood index × % Time with SNAP

−1.19

(3.23)

9.30

(3.30)**

5.36

(2.79)

4.05

(2.62)

 % Time with SNAP

−1.17

(5.84)

11.72

(5.34)*

10.75

(3.74)**

8.48

(3.31)**

 % Time with SNAP and TANF

2.01

(5.28)

2.92

(4.42)

1.54

(3.00)

0.05

(2.59)

 % Time with high-income nonreceipt

−4.77

(3.92)

3.15

(3.41)

1.86

(2.41)

2.26

(2.12)

 Female × Neighborhood index

0.07

(0.59)

−0.42

(0.54)

0.28

(0.49)

−0.33

(0.47)

 Female × % Time with SNAP × Neighborhood index

0.68

(2.86)

1.42

(2.60)

−2.97

(2.28)

−1.37

(2.27)

 Female × % Time with SNAP

4.51

(4.80)

0.51

(4.07)

−3.40

(3.15)

−1.23

(2.86)

 Female × % Time with SNAP and TANF

4.78

(2.84)

1.92

(2.17)

2.90

(1.74)

1.15

(1.56)

 Female × % Time with high-income nonreceipt

2.24

(2.28)

1.07

(1.90)

−0.52

(1.41)

−0.39

(1.26)

N

1,661

2,212

2,924

3,211

Notes: For the childhood variable-only results, all childhood covariates were included in the regression models. For the child and adult variable models, all childhood and adult covariates were included in the regression models.

p < .10; *p < .05; **p < .01; ***p < .001 (two-tailed tests)

In Table 7 and Figs. 3 and 4, we present predicted adult BMI values by age group, neighborhood condition, and SNAP usage. As expected, SNAP usage during preschool years has little effect on adult BMI. However, for the school-age group, adult BMI increases dramatically as SNAP usage increases, but primarily for those living in more-advantaged neighborhoods. We find similar but less-dramatic results for SNAP use across the two oldest age ranges. Figures 3 and 4 tell a striking story: those growing up in disadvantaged neighborhoods have fairly consistent adult BMI predictions regardless of level of SNAP receipt. By contrast, those growing up in advantaged neighborhoods see predicted adult BMI increases that rise quickly to the level of obesity, particularly when receipt occurs during primary school years.
Table 7

Weighted adult BMI predictions, using neighborhood index

 

Assumes Low Income but No SNAP Use as a Child or Adult

Assumes 25% Use of SNAP as a Child, and $617/Yearly as an Adult

Assumes 43% Use of SNAP as a Child, and $1,443/Yearly as an Adult

Childhood-Only Variables

 Ages 0–4

 +1 SD

29.16

29.39

29.55

 −1 SD

30.32

30.89

31.30

 Ages 5–8

 +1 SD

23.86

29.41

33.40

 −1 SD

25.25

25.73

26.08

 Ages 9–13

 +1 SD

26.80

30.01

32.31

 −1 SD

27.13

28.47

29.44

 Ages 14–18

 +1 SD

25.88

28.60

30.56

 −1 SD

27.14

28.31

29.14

Child and Adult Variables

 Ages 0–4

 +1 SD

29.13

30.85

33.14

 −1 SD

30.56

31.99

33.74

 Ages 5–8

 +1 SD

23.84

30.20

35.49

 −1 SD

26.26

27.24

28.46

 Ages 9–13

 +1 SD

25.92

30.10

33.70

 −1 SD

27.67

29.79

31.74

 Ages 14–18

 +1 SD

25.51

29.00

32.01

 −1 SD

27.56

29.41

31.14

Notes: Predictions are derived from the regression models presented in Table 6. For the neighborhood index, higher values indicate more-advantaged neighborhoods.

https://static-content.springer.com/image/art%3A10.1007%2Fs13524-012-0115-y/MediaObjects/13524_2012_115_Fig3_HTML.gif
Fig. 3

Predicted adult BMI by age of receipt for those growing up in advantaged neighborhoods (Table 7)

https://static-content.springer.com/image/art%3A10.1007%2Fs13524-012-0115-y/MediaObjects/13524_2012_115_Fig4_HTML.gif
Fig. 4

Predicted adult BMI by age of receipt for those growing up in disadvantaged neighborhoods (Table 7)

Discussion and Conclusion

At present, the United States has a complex, somewhat contradictory, and certainly public relationship with consumption. On the one hand, the Obama administration has recognized childhood obesity as a national crisis and proposed to reduce obesity in children from 20% to 5% by 2030. On the other, the National Center for Children in Poverty (NCCP) recently publicized disturbing figures on food insecurity: 21% of U.S. families with children report at least one family member with decreased or disrupted eating patterns because of financial hardship (Wight et al. 2010). Of course, consumption is only one component in a complex equation of factors that determine weight and health, but it remains an important and at times controversial focus of U.S. food assistance policy.

Our study seeks to clarify the relationship between one food assistance program (the Supplemental Nutrition Assistance Program (SNAP)) and one indicator of health and well-being (BMI scores) in the context of residential conditions. To date, there has been mixed evidence for the relationship between SNAP use and BMI; studies have generally shown that SNAP use has either no effect or a positive effect on BMI or the likelihood of obesity, particularly for women. One source of sometimes conflicting results for this type of work is that low-income families who receive SNAP income may be different from those who do not in ways that confound comparisons between the two groups. We use sibling fixed-effects models and proportions of time with receipt to decrease any bias that may be due to the choices that people make in using SNAP assistance.

We find that for those families with some SNAP income, almost all have within-family differences in time using SNAP during childhood. Similarly, nearly 60% of sibling groups have within-family differences in time with low income without using SNAP or TANF—and, for those families who spend some time in this state, almost all siblings groups have within-family differences. We find that OLS models with only sibling observations tend to underestimate the effects of both SNAP usage and the interaction of SNAP usage and neighborhood conditions during childhood relative to OLS models where all children (not just siblings) are included. Thus, our findings may be viewed as conservative estimates of the relationships that we examine.

Controlling for a host of other variables, we find that childhood SNAP receipt has a positive effect on adult BMI when compared with a condition of low-income nonreceipt. Moreover, although neither childhood nor adult neighborhood characteristics directly affects adult BMI values, SNAP usage during childhood has a strong and significant effect on adult BMI, but primarily and to a larger extent for children growing up in advantaged neighborhoods. These effects are particularly pronounced for boys.

We find no evidence that either a measure of relative deprivation or the county crime rate have an effect on adult BMI, either directly or in interaction with SNAP assistance. However, in models using these neighborhood or county indicators, the direct effects of SNAP receipt are notably weaker, although still positive, relatively large, and significant. This suggests that models of SNAP receipt are sensitive to the inclusion of residential indicators. Given a body of work suggesting no or weak direct relationships between SNAP receipt and obesity (e.g., Ver Ploeg et al. 2007), further exploration of this dynamic is needed.

Similar to Duncan et al. (1998), Levy and Duncan (2000), and Vartanian and Buck (2005), we find that factors at the beginning years of life have sizable and significant consequences in adulthood. At preschool age (0 to 4), spending time in a recipient family relative to a low-income, nonrecipient family has no effect on adult BMI (although in the interaction-by-sex model, there is some suggestion that spending one’s preschool years in a non-eligible, nonrecipient family leads to substantially lower adult BMI levels) (see Table 6). However, at older ages, particularly at school age (5 to 8), children are affected by time receiving SNAP income, with more time spent receiving such aid increasing adult BMI, especially for those growing up in advantaged areas.

Amidst concerns of an obesity epidemic, it is tempting to construe "higher adult BMI" findings as strictly negative in a prescriptive sense. However, it is possible that higher average adult BMI findings for children of low-income, recipient families, particularly when compared with low-income, nonrecipient families, mean simply that SNAP is doing its job. Adult weight is not being compromised by food insecurity or undernutrition in childhood.

Still, our findings suggest that predicted adult BMI values for children growing up in low-income families are in the overweight range (BMI: 25–29.9), regardless of residence in above- or below-average neighborhoods, and that SNAP assistance may push these predicted values into obese range (BMI: greater than 30). Moreover, although predicted adult BMI values for low-income, nonrecipient children tend to be lower if they grow up in above-average neighborhoods, the trend reverses itself under conditions of increasing SNAP receipt. In every model presented in Tables 5 and 7, predicted adult BMI in above-average neighborhoods is higher than in below-average neighborhoods for those children with average or above-average SNAP use in childhood and adulthood.

A potential explanation of our findings is that exposure to a greater number of relatively inexpensive food sources may not necessarily lead to greater consumption of healthy foods, but simply to greater consumption altogether. Families that receive SNAP funds have low incomes, and thus may quite reasonably purchase as much as they can with those limited incomes. Thus, foods higher in fat or caloric content, or foods that are easier to prepare, may still be the best food choices when the price of healthier foods, although less expensive than in corner grocery markets, is still high. Moreover, even though we do not find any support for the effect of relative deprivation when understood as a “moving target” (i.e., for the effect of having a lower income than one’s neighbors), it remains entirely possible that growing up as a child in a low-income family amidst an absolute degree of affluence has an isolating and stigmatizing effect that is compounded by the use of food assistance (Galea and Ahern 2005; Sobal 1991).

Another possibility is that in this case, SNAP receipt is a proxy for a level of severe and/or sustained hardship not found among low-income families with no receipt. Keep in mind, however, that our fixed-effects regression models compare children who grow up in the same families but under sometimes very different circumstances, including differing amounts of time with SNAP receipt. As evidenced by predicted adult BMI values that almost uniformly meet the criteria for overweight (Table 5; Fig. 1), children in low-income households are not faring particularly well, even under “optimal” conditions of nonreceipt and residence in above-average neighborhoods.

We have several important caveats to our study. First, we do not have specific residential neighborhood data, such as the number of food sources. We rely instead on studies showing that residents of low-income and otherwise disadvantaged neighborhoods tend to have less access to and less variety in food sources (e.g., Powell et al. 2007). Second, the crime data used are at the county level, rather than the neighborhood level. Thus, the effects of higher-crime neighborhoods within lower-crime counties (and vice versa) may be masking particular neighborhood effects in our analyses. Lastly, the lack of usable information on individual-level physical activity or energy expenditure in the study data set prevents us from incorporating an important predictor of BMI in our models.

Several policy prescriptions may be drawn from this study. Indeed, a number of state politicians, and perhaps with particular fervor of late, have proposed disallowing SNAP recipients to use their benefits to purchase specific high-calorie, low-nutrient food items, such as sodas and sugar-sweetened foods (Miranda 2011; Replogle 2010). To date, however, such proposals have been unable to overcome states’ unwillingness to expend the money or time needed to “police” dietary choice; the reluctance of the average citizen to enforce restrictions on SNAP recipients that are not applied to the general population; and, in the case of a 2004 petition from the state of Minnesota, rejection by the federal government (Replogle 2010).

An alternative approach would be to use the SNAP to close the price gap between fresh and processed foods by placing an additional subsidy on foods that are high in nutrient value. Decreasing the relative price of nutrient-rich foods may increase the likelihood of purchasing such foods. This will not, however, assist those with low incomes who do not receive SNAP funds, nor will it alter long-standing dietary practices, to the extent that they exist. To this point, Bhattacharya and Currie (2000) found that nutrient deficiencies in young adults appear not to result from food shortages or general resource constraints. There is some evidence that in addition to access to high-quality foods and the resources necessary to purchase them, individuals benefit from information about both dietary practice and food-related behaviors (Cason et al. 2002; Lytle et al. 1996). Even a recent article, questioning the economic rationale for public health efforts to curb adult obesity, acknowledges the value of an educative and perhaps even regulatory public health approach targeted to children and youth (Finkelstein et al. 2004).

Proposals either to ban or to incentivize the purchase of certain foods with SNAP funds through tax or subsidy plans may be used to reify the idea that the problems of obesity and overweight still come down to individual intake. Our findings suggest that the reality is substantially more complex. Because consumption and purchasing practices are not explicitly included in our models, we cannot assess the relative contributions of these factors on adult weight. However, the strong and persistent influence of childhood residential conditions on the relationship between childhood SNAP use and adult weight belies simplistic assumptions of causality between individual “choice” and weight or obesity. Assisting both SNAP-recipient and nonrecipient families—and, indeed, all U.S. families—to promote the long-term health of their children will require a rethinking of many aspects of food cost, access, and education.

Acknowledgment

This article was funded by a grant from the Economic Research Service at the U.S. Department of Agriculture.

Copyright information

© Population Association of America 2012