- Email: [email protected]
Supplemental nutrition assistance program and body weight outcomes: The role of economic contextual factors
Social Science & Medicine 74 (2012) 1874e1881
Contents lists available at SciVerse ScienceDirect
Social Science & Medicine journal homepage: www.elsevier.com/locate/socscimed
Supplemental nutrition assistance program and body weight outcomes: The role of economic contextual factors Euna Han a, *, Lisa M. Powell b, Zeynep Isgor c a
Gachon University, College of Pharmacy, 534-2 Yeonsu-Dong, Yeonsu-Gu, Inchon, 406-799, South Korea Department of Economics, Institute for Health Research and Policy, University of Illinois at Chicago, 1747 West Roosevelt Road, M/C 275, Chicago, IL 60608, USA c Department of Economics, University of Illinois at Chicago, 1747 West Roosevelt Road, M/C 275, Chicago, IL 60608, USA b
a r t i c l e i n f o
a b s t r a c t
Article history: Available online 20 March 2012
We explored the extent to which economic contextual factors moderated the association of Supplemental Nutrition Assistance Program (SNAP) participation with body mass index (BMI) among lowincome adults whose family income (adjusted for family size) is less than 130% of the federal poverty guideline. We drew on individual-level data from the Panel Study of Income Dynamics in the United States, including three waves of data in 1999, 2001, and 2003. Economic contextual data were drawn from the American Chamber of Commerce Researchers Association for food prices and Dun & Bradstreet for food outlet measures. In addition to cross-sectional estimation, a longitudinal individual fixed effects model was used to control for permanent unobserved individual heterogeneity. Our study found a statistically significant joint moderating effect of the economic contextual factors in longitudinal individual fixed effects model for both women (BMI only) and men (both BMI and obesity). For both women and men, SNAP participants’ BMI was statistically significantly lower if they faced increased numbers of available supermarkets/grocery stores in the longitudinal model. A simulated 20% reduction in the price of fruits and vegetables resulted in a larger decrease in BMI among SNAP participants than non-participants for women and men, whereas a simulated 20% increase in the availability of supermarkets and grocery stores resulted in a statistically significant difference in the change in BMI by SNAP participation for women but not for men. Policies related to economic contextual factors, such as subsidies for fruits and vegetables or those that would improve access to supermarkets and grocery stores may enhance the relationship between SNAP participation and body mass outcomes among food assistance program participants. Ó 2012 Elsevier Ltd. All rights reserved.
Keywords: Supplemental nutrition assistance program (SNAP) Body mass index (BMI) Economic contextual factors United States
Introduction The prevalence of obesity in the U.S. has risen at a striking rate over the past few decades, with one-third of American adults reported to be obese in 2007e08 (Flegal, Carroll, Ogden, & Curtin, 2010). Particularly, low-socioeconomic status (SES) individuals have shown higher prevalence of obesity (Chang & Lauderdale, 2005; Rhoades, Altman, & Cornelius, 2004; Stunkard & Sorensen, 1993; Townsend, Peerson, Love, Achterberg, & Murphy, 2001). Such phenomena have drawn a tremendous interest in understanding the contributors to this epidemic given the adverse associations of obesity with numerous health outcomes (USDHHS,
* Corresponding author. Tel.: þ82 32 820 4766; fax: þ82 32 820 4829. E-mail addresses: [email protected], [email protected] (E. Han), [email protected] (L.M. Powell), [email protected] (Z. Isgor). 0277-9536/$ e see front matter Ó 2012 Elsevier Ltd. All rights reserved. doi:10.1016/j.socscimed.2012.02.032
2001) and non-health outcomes such as wages (Han, Norton, Stearns, 2009; Han, Norton, & Powell, 2011), and the mounting societal burden from the high medical costs associated with obesity (Finkelstein, Trogdon, Cohen, & Dietz, 2009). Several previous studies have examined the Supplemental Nutrition Assistance Program (SNAP), previously called the Food Stamp Program, as one potential channel linked to high body weight outcomes for its eligible population (Baum, 2007; Chen, Yen, & Eastwood, 2005; Gibson, 2003, 2006; Meyerhoefer & Pylypchuk, 2008; Townsend et al., 2001; Zargosky & Smith, 2009). As the largest domestic food and nutrition assistance program in the U.S., the SNAP is one of the most important programs to combat food insecurity (Ver Ploeg & Ralston, 2008). Nonetheless, there have been ongoing debates on whether the provision of food vouchers has potentially encouraged its recipients to consume more food compared to cash assistance (Besharov, 2002; Fox, Hamilton, & Lin, 2004; Fraker, 1990; Fraker, Martini, &
E. Han et al. / Social Science & Medicine 74 (2012) 1874e1881
Ohls, 1995). Given that the SNAP benefits are entirely federally funded (Office of Analysis, 2001) and increased body weight puts substantial burden on the public (Finkelstein et al., 2009; USDHHS, 2001), it is important to explore the characteristics of the relationship of SNAP participation with body mass and effective policy measures to potentially complement the relationship so that the SNAP can improve participants’ body weight. An increasing number of studies have highlighted the importance of economic contextual factors such as food prices and food store outlet densities as important environmental factors that may contribute to the obesity epidemic. Higher fast food prices and lower grocery food prices (Chou, Grossman, & Saffer, 2004), higher prices of sugar (Miljkovic, Nganje, & de Chastenet, 2008), and lower fruit and vegetable prices (Beydoun, Powell, & Wang, 2008; Powell & Han, 2011) have been associated with lower body mass index (BMI) and obesity prevalence among adults. The geographic availability of foods through food stores also has been linked to the obesity epidemic. The greater availability of supermarkets has been associated with more fruit and vegetable intake, more healthful diets, and lower rates of obesity among American adults (Morland, Diez-Roux, & Wing, 2006; Morland, Wing, Diez-Roux, & Poole, 2002). This study built on the previous literature and examined the association of SNAP participation with body weight outcomes among people in low SES who are potential SNAP recipients, with a focus on whether economic contextual factors moderated the relationship between SNAP participation and body weight outcomes. Not only did we measure the association of the moderating effects of the economic contextual factors in a cross-sectional setting, we also estimated longitudinal individual-level fixed effects models using data from the Panel Study of Income Dynamics (PSID) in the United States to control for unobserved permanent individual heterogeneity. Methods Empirical models We regressed individual body weight outcomes (either linear BMI or a non-linear measure of obesity) on SNAP participation, economic contextual factors, and their interactions to measure the extent to which the economic contextual factors moderated the association of SNAP participation with body weight outcomes. We measured the economic contextual factors via the monetary cost of foods as well as opportunity cost of obtaining foods. The measurement of the monetary cost of foods included fruit and vegetable prices (proxy for healthy foods) and fast food prices (proxy for unhealthy foods). The opportunity cost of food consumption was measured by the geographic accessibility of food outlets including supermarkets and grocery stores. We controlled for neighborhood median household income as a proxy for neighborhood socioeconomic patterns. Our basic estimation was: SNAP
BMIist ¼ b0 þ b
SNAPPRICE
þb
þb
PRICE
SNAPist1 þ b
SNAPOC
PRICEst1 OC
SNAPist1 PRICEst1 þ b
OCst1
X
SNAPist1 OCst1 þ b Xist þ mi þ 3 ist (1)
where the subscripts i, t, and s denoted individual, year, and zip code, respectively. The b’s were parameters to be estimated. m was time invariant unobserved individual heterogeneity and 3 was a time-varying error term that was assumed to be normally distributed. Cross-sectional estimation does not control for the permanent portion of the error term, m, whereas the longitudinal
1875
individual fixed effects model controls for the time-constant unobserved individual heterogeneity. BMI was a linear measure of body mass calculated as weight in kilograms divided by height in meters squared. SNAP denoted SNAP participation. PRICE was a vector that measured the prices of fruits and vegetables and fast food in zip code s in year t 1. OC was a vector that included measures of the availability (per 10,000 capita per 10 square miles) of supermarkets and grocery stores in zip code s in year t 1. Xist was a vector of individual and household characteristics including year dummy variables. In the vector Xist, we also controlled for zip code level median household income and the price match quality measure of the distance in miles between the centroid of the zip code and the closest city in the price data. We reported the marginal effect of SNAP (bSNAP þ bSNAP PRICEPRICEst þ bSNAP OCOCst) which was calculated at the mean value of each economic contextual variable and all other covariates. We also reported the marginal effects of the economic contextual factors for both non-SNAP and SNAP recipients. The coefficients of the economic contextual variables (bPRICE and bOC) represent the marginal effect of a unit increase in those variables on BMI among non-SNAP participants. The marginal effect of a unit increase of each economic contextual variable on BMI among SNAP participants is the combined coefficients of an economic contextual variable and its interaction with SNAP (bPRICE þ bFSP PRICE and bOC þ bFSP OC). All the estimations were run separately by gender. In addition, we conducted an F-test to explore whether the moderating effects of the economic contextual variables were jointly statistically significant. We conducted simulations to assess whether the effects of policies related to the economic contextual variables on BMI differed between participants and non-participants in SNAP. For the simulations, we obtained differences in BMI before and after any simulated changes in economic contextual variables for participants and non-participants. We undertook simulations based on participation and non-participation and in each case reduced the price of fruits and vegetables by 20% or increased the number of supermarkets and grocery stores by 20%. Finally, we calculated the difference-in-difference as participants’ after and before simulated policy change difference minus the non-participants’ after and before simulated policy change difference. All standard errors were bootstrapped. This study was approved by the Institutional Review Board (IRB) of the University of Illinois at Chicago. Data Individual-level data We used the PSID in the United States, which began in 1968 as a longitudinal study of a representative sample of U.S. individuals and their co-residents. The PSID interviewed individuals from families in the core sample annually from 1968 to 1996, and biennially since 1997. The estimation sample for this study consisted of individuals whose per capita family income (adjusted for family size) was less than or equal to 130% of the federal poverty guideline. Weight status related measures became available in the PSID in 1999. Our estimation sample included three waves of data starting in 1999 and onwards (1999, 2001, and 2003). The final estimation samples of 2,391 women and 1,351 men were obtained after restricting the sample to those who were at or below 130% of the federal poverty line, between the ages of 18e65 years, not pregnant at the time of interview, and with non-missing information for any of the variables in the estimations.
1876
E. Han et al. / Social Science & Medicine 74 (2012) 1874e1881
Our dependent variable was BMI, measured as self-reported weight in kilograms divided by self-reported height in meters squared. Several previous studies, including Cawley (2004), attempted to adjust potential measurement errors for self-reported height and weight using the third National Health and Nutrition Examination Survey (NHANES III). However, we did not do so because the size or magnitude of the errors in self-reported height and weight in the NHANES III may be different than those in the PSID; in particular, the respondents in the NHANES III were aware that their weight and height would be measured after their selfreports of weight and height (Han, Norton, & Stearns, 2009; USDHHS, 1996). We further classified BMI into a clinical weight category of obese (BMI 30) as an additional dependent variable to investigate whether differential patterns of the moderating effects of economic contextual factors on the effect of SNAP participation on BMI were found at the right tail of the BMI distribution. SNAP participation, the primary independent variable of interest, was defined as a dichotomous indicator of whether the respondents or their spouses reported participating in the SNAP during the previous calendar year from the year of interview. Other covariates included the following: race/ethnicity (African American, Hispanic, other, with White as the reference), age, marital status (never married, separated, divorced, widowed, with married as the reference), number of children, education level (high school, some college, college or more, missing indicator for the education level, with less than high school as the reference), continuous measure of per capita family income, and year of the interview wave (2001 and 2003, with 1999 as the reference). We also controlled for the degree of urbanization of the individuals’ zip code of residence based on the Census 2000 data that measured population size within a zip code inside urbanized areas, outside urbanized areas (referred to as suburban areas), and in rural areas. Zip code level median household income based on the Census 2000 data was controlled for. Table 1 presents descriptive statistics of the estimation sample. The average BMI was 28.5 for women and 27.6 for men, and 35% and 26% of women and men, respectively, were classified as obese. Approximately 42% and 26% of women and men participated in SNAP during the previous calendar year, respectively. The average age of the sample persons was 38 years. Among women and men, respectively, approximately 29% and 39% were non-Hispanic white, 58% and 43% were non-Hispanic black, and 10% and 14% were Hispanic. Approximately one quarter of women (28%) and one half of men (47%) were married at the time of interview. Approximately one fourth of the sample population lived in a rural area for both genders. The approved amount for federal funding for the Food Stamp Nutrition Education Program (FSNE) per capita at state level was on average $0.3 in 1982e1984 dollars. Contextual program control We controlled for the approved amount of federal funding for the FSNE at the state level on a per capita basis as a proxy measure for unobserved state level effects of providing nutrition education to individuals that may increase knowledge and awareness regarding food intake. The FSNE is an important outreach program for low-income families and youth to adopt healthy lifestyles, such as consumption of more fruits, vegetables, whole grains, and lowfat dairy products, and engagement in daily physical activity. Between 1992 and 2004, approved federal funding that reimburses half of a state’s allowable costs increased from less than $1 million to $229 million (Office of Analysis, 2001). Economic contextual data Food price data were based on the American Chamber of Commerce Researchers Association (ACCRA) Cost of Living Index
Table 1 Summary statistics.
Dependent variables: BMI Obese Independent variables of interest SNAP participation Price of fruits and vegetables ($1982e84) Price of fast food ($1982e84) # of supermarkets and grocery storesb Average distance of price match (miles) Approved federal funding for Food Stamp Nutrition Education Program per capita state population ($1982e84) Age Race: Whitea Race: Black Race: Hispanic Race: Other race Marital status: Marrieda Marital status: Never married Marital status: Widowed Marital status: Divorced Marital status: Separated Number of children Education: Less than high schoola Education: High school Education: Some college Education: College or more Per capita family income ($1982e84) (`000) Median household income ($2000) (`000)c Urbanicity: Urbana Urbanicity: Suburban Urbanicity: Rural/Farm Year: 1999a Year: 2001 Year: 2003 N
Women
Men
All
All
28.4864 (7.2091) 0.3484
27.5756 (5.7316) 0.2591
0.4245 0.7783 (0.1062)
0.2554 0.7757(0.1124)
2.6941 (0.1595) 10.6383 (55.3120) 24.0131 (22.3081) 0.2973 (0.2778)
2.7177(0.1675) 8.7490 (27.9155) 26.6061 (24.1971) 0.3144 (0.2973)
37.6014 (11.3335) 0.2865 0.5826 0.1008 0.0301 0.2769 0.4040 0.0435 0.1706 0.1050 1.8076 (1.5464) 0.4262 0.3254 0.1547 0.0410 2.2980 (1.4200)
38.2258 (11.7433) 0.3901 0.4308 0.1429 0.0363 0.4730 0.3331 0.0126 0.1207 0.0607 1.3294 (1.5612) 0.4330 0.3175 0.1399 0.0725 2.5612 (1.6721)
33.2029 (10.6081) 35.0075 (12.0591) 0.6491 0.1397 0.2112 0.3446 0.3074 0.3480 2391
0.5892 0.1665 0.2443 0.3486 0.3020 0.3494 1351
Notes: Summary statistics are not weighted. Standard deviations (SD) are shown in parenthesis for continuous variables. a Denotes reference categories in regression models. Although the summary statistics for the missing indicator for the education variable and the age squared term are not included in the table, all regressions included these variables as control measures. b Food store related facility variable is defined per 10,000 capita per 10 square miles. c Median household income is based on Census 2000 Data and at the zip code level.
reports. These reports contain quarterly information on prices across more than 300 US cities and have been used in a number of previous studies (Auld & Powell, 2009; Chou et al., 2004; Powell et al., 2011; Powell & Chaloupka, 2009; Sturm & Datar, 2005). The price data were matched to the PSID sample based on the closest city match, which was determined by the shortest straight line distance between the centroid point of the adult’s zip code and the centroid point of the ACCRA price cities. We controlled for the price match quality using a variable based on this distance in miles in all estimations. We created the following two food price indices from the items provided in the ACCRA: 1) a fruit and vegetable price index as a weighted sum of the price of potatoes, bananas, lettuce, canned sweet peas, canned peaches, and frozen corn and 2) a fast food price index as a weighted sum of the prices of a McDonald’s QuarterPounder with cheese, a thin crust regular cheese pizza at Pizza Hut and/or Pizza Inn, and fried chicken (thigh and drumstick) at
E. Han et al. / Social Science & Medicine 74 (2012) 1874e1881
Kentucky Fried Chicken and/or Church’s Fried Chicken. Each of the price indices was weighted based on expenditure shares derived from the Bureau of Labor Statistics (BLS) Consumer Expenditure Survey. All prices were deflated by the BLS Consumer Price Index (1982e1984 ¼ 1). The average real price of fruits and vegetables and a fast food meal were $0.78 and $2.70, respectively, in our sample (Table 1). Data on food store outlets were drawn from a business list developed by Dun and Bradstreet (D&B) (Dun and Bradstreet, 2005). The outlet density data were matched to the PSID by year at the zip code level. We adjusted the number of available outlets per 10,000 capita per 10 square miles using the Census 2000 zip code level population and land area estimates in order to account for accessibility both in terms of congestion (per capita) and distance (per land area). We measured the number of supermarkets and grocery stores based on the 6-digit primary Standard Industrial Codes. There were on average 9e10 supermarkets/grocery stores per 10,000 capita per 10 squares miles per zip code with slight variations by gender (Table 1). All price and food outlet density measures were included in the estimation as one year lagged variables given that SNAP participation, the variable of interest, represented receipt of SNAP in the previous calendar year. Results BMI outcome The results indicated that female SNAP participants were estimated to have 1.31 units higher BMI than female non-participants
1877
when all covariates in the base model were set to their mean values in the cross-sectional model. We also rejected the null hypothesis of no joint moderating effects of the economic contextual variables (p-value ¼ 0.0122). An additional supermarket/grocery store in the area was estimated to decrease BMI by 0.0401 units for SNAP participants whereas for non-participants the corresponding estimate was 0.0029 and statistically insignificant. Controlling for permanent unobserved individual heterogeneity in the individual fixed effects model for women, we found no statistically significant relationship between SNAP participation and BMI. The economic contextual factors had joint moderating effects that increased the extent of the rewarding effect of SNAP participation on BMI among women given that the null hypothesis of no joint moderating effects was rejected (p-value ¼ 0.0147). Particularly, the rewarding effect of the number of supermarkets/ grocery stores remained statistically significant in the individual fixed effects model among SNAP participants (one additional supermarket/grocery store in the areas was associated with 0.0376 lower BMI units) whereas no such statistically significant effect was found for non-participants (see Table 2). For men, no statistically significant association of SNAP participation with BMI was found across all models. In the cross-sectional model, we found no statistically significant effects of economic contextual variables on BMI among either participants or nonparticipants in SNAP. In the individual fixed effects model controlling for time invariant unobserved individual heterogeneity, the moderating effect of the number of supermarkets/grocery stores was statistically significant for men. The number of supermarkets statistically significantly reduced BMI for both SNAP participants and non-
Table 2 Estimated regression coefficients from cross-sectional and longitudinal individual fixed effects models of BMI, by gender. Outcome variable: BMI
Price of fruits and vegetables Price of fast food # of supermarkets and grocery stores SNAP participation Price of fruits and vegetables SNAP participation Price of fast food SNAP participation # of supermarkets and grocery stores SNAP participation Funding for FSNE Program per capita state population ($1982e84) Median household income ($2000) (`000) Suburban Rural/farm Age Black Hispanic Other race Never married Widowed Divorced Separated Number of children Education: High school Education: Some college Education: College or more Per capita family income ($1982e84) (`000) Year: 2001 Year: 2003 Constant Marginal effect of SNAP participation P-value for joint F-test of all the moderating effects
Women (N ¼ 2391)
Men (N ¼ 1351)
Cross-sectional
Individual FE
Cross-sectional
Individual FE
2.1544 1.6637 0.0029 8.5028 1.3891 3.3896 0.0372*** 0.3023 0.0373** 0.2670 0.9741 0.1767 2.8132*** 0.8171 1.0624 1.0052** 0.8835 0.6073 0.2243 0.1687 0.2320 0.6656 0.4869 0.1165 0.0607 0.0901 25.042*** 1.3147*** 0.0122
2.1907 0.0786 0.0018 0.5536 0.1821 0.1894 0.0358*** 0.6845 0.0210 1.0915 0.9475 e e e e 0.4423 1.1392 0.4856 0.3192 0.3569** 0.3644 0.3433 2.3659 0.0471 0.5311** 0.7483** 24.466*** 0.2823 0.0147
0.6808 0.8005 0.0050 9.0352 7.0760 1.5535 0.0090 0.6054 0.0133 0.3099 0.2035 0.2994*** 0.8728 0.0001 0.7547 0.1838 1.0061 0.2782 0.1658 0.1841 0.8881 0.2900 0.3433 0.1049 0.4076 0.4263 21.007*** 0.5972 0.3490
2.4910 2.1917 0.0114** 16.052*** 2.9742 5.1559*** 0.0213 0.3071 0.0225 0.8452 0.4648 e e e e 0.9704 e 0.8053 1.6696 0.1510 0.1523 2.6838** 1.5868 0.1838** 0.6199** 1.0798** 35.199*** 0.0815 0.0312
(2.2222) (1.4203) (0.0016) (5.5008) (2.7434) (1.9035) (0.0129) (0.7085) (0.0185) (0.6588) (0.5444) (0.1100) (0.4725) (0.7544) (0.8782) (0.4901) (0.9635) (0.5704) (0.6836) (0.1423) (0.4449) (0.5164) (0.7544) (0.1144) (0.3420) (0.4516) (4.2949) (0.4073)
(1.8434) (1.1671) (0.0047) (4.2141) (1.9192) (1.4270) (0.0110) (0.6464) (0.0170) (0.5785) (0.6147)
(0.4316) (1.0083) (0.5714) (0.5583) (0.1583) (0.6601) (0.6578) (1.4757) (0.0860) (0.2221) (0.3381) (3.3388) (0.3065)
(2.5856) (1.2926) (0.0037) (7.2030) (4.0452) (2.2840) (0.0179) (0.6349) (0.0162) (0.4754) (0.5099) (0.0974) (0.4816) (0.6461) (0.6982) (0.4706) (2.1504) (0.6880) (0.9750) (0.1519) (0.4806) (0.5100) (0.7039) (0.0890) (0.3899) (0.5136) (4.2988) (0.4494)
(3.4805) (1.9478) (0.0050) (5.9614) (2.8090) (1.9613) (0.0154) (0.4978) (0.0151) (1.0192) (0.8162)
(0.6713) (0.8149) (0.9745) (0.2388) (0.8009) (1.1917) (1.6377) (0.0896) (0.2835) (0.4611) (6.8372) (0.3234)
Notes: All models included SNAP participation, two food price indices (fruit and vegetable and fast food meal), one food outlet density measure (the number of supermarkets/ grocery stores), and the interaction terms of SNAP participation with the food price and food outlet density variables as the main variables of interest. Match quality variable (distance from the centroid of zip code to the centroid of the nearest ACCRA city), age squared, and education missing indicator were all controlled for in the models but corresponding coefficients were not shown. Standard errors are in parentheses and were adjusted for clustering at the zip code level for cross-sectional and individual fixed effects models. **p 0.05, ***p 0.01.
1878
E. Han et al. / Social Science & Medicine 74 (2012) 1874e1881
participants with a bigger effect for participants. Specifically, one additional supermarket/grocery store was estimated to decrease BMI by 0.0114 units among non-participants in SNAP whereas the size of the decrease was estimated to be 0.0326 units for participants. Further, the null hypothesis of no joint moderating effects of the economic contextual variables was rejected (p-value ¼ 0.0312) (see Table 2). For other covariates included in all models, the signs of the coefficients were as we expected in general. Non-Hispanic black women had higher BMI by 2.8 unites than non-Hispanic white women. Women with more children had higher BMI when unobserved individual heterogeneity was controlled for in the longitudinal model. There was an upward time trend of average BMI for both genders to the extent that the average BMI increased by approximately one unit in year 2003 compared to year 1999.
would decrease the likelihood of obesity by 0.03 and 0.19 percentage points for non-participants and participants in SNAP, respectively, in the cross-sectional model. The effect of the number of supermarkets/grocery stores on the likelihood of obesity turned statistically insignificant in the longitudinal individual fixed effects model (Table 3). For men, one more supermarket/grocery store decreased the likelihood of obesity by 0.08 percentage points for only nonparticipants in SNAP in the cross-sectional model. Further controlling for the permanent individual heterogeneity in the longitudinal individual fixed effects model revealed no statistically significant effects of the economic contextual variables regardless of SNAP participation. The results also showed that the null hypothesis of no joint moderating effect was rejected in the individual fixed effects model (p-value ¼ 0.0132). The marginal effect of SNAP participation on the likelihood of obesity was negative but statistically insignificant.
Obesity outcome SNAP participant women showed higher likelihood of being obese by 7.3 percentage points in the cross-sectional model when all other covariates were set to their mean values. The economic contextual factors jointly reduced the penalizing effect of SNAP participation on the likelihood of obesity in the cross-sectional model for women (p-value ¼ 0.0073 for the null hypothesis of no joint moderating effects). The joint moderating effect was no longer statistically significant when time invariant individual heterogeneity was further controlled for in the individual fixed effects model. We also found a small negative (rewarding) effect of the number of supermarkets/grocery stores for the likelihood of obesity to the extent that one more supermarket/grocery store
Sensitivity analyses We performed sensitivity analyses by replicating all estimations including only the price variables as economic contextual measures to ensure our main findings. This was to test whether the estimation results with only food price variables were comparable to when the arguably endogenous outlet density variables were included. The estimates for SNAP participation, food prices, and the interaction terms of SNAP participation and food prices overall remained similar in terms of direction, magnitude, and statistical significance between those two specifications (results not shown in Tables).
Table 3 Estimated regression coefficients from cross-sectional and longitudinal individual fixed effects models of the probability of obesity, by gender. Outcome variable: OBESE
Women (N ¼ 2391) Cross-sectional
Price of fruits and vegetables Price of fast food # of supermarkets and grocery stores SNAP participation Price of fruits and vegetables SNAP participation Price of fast food SNAP participation # of supermarkets and grocery stores SNAP participation Funding for FSNE Program per capita state population ($1982e84) Median household income ($2000) (`000) Suburban Rural/farm Age Black Hispanic Other race Never married Widowed Divorced Separated Number of children Education: High school Education: Some college Education: College or more Per capita family income ($1982e84) (`000) Year: 2001 Year: 2003 Constant Marginal effect of SNAP participation P-value for joint F-test of all the moderating effects
0.0741 0.1122 0.0003*** 0.8080** 0.0523 0.3184** 0.0017*** 0.0437 0.0016 0.0739 0.1015*** 0.0056 0.1688*** 0.0067 0.0627 0.0625 0.0972 0.0327 0.0493 0.0099 0.0122 0.0048 0.0249 0.0054 0.0059 0.0050 0.3171 0.0728*** 0.0073
(0.1561) (0.1019) (0.0001) (0.3922) (0.2035) (0.1365) (0.0006) (0.0521) (0.0011) (0.0444) (0.0377) (0.0068) (0.0328) (0.0479) (0.0598) (0.0329) (0.0588) (0.0347) (0.0426) (0.0091) (0.0275) (0.0351) (0.0509) (0.0080) (0.0232) (0.0313) (0.2951) (0.0256)
Men (N ¼ 1351) Individual FE 0.0192 0.0217 0.0004 0.0700 0.0302 0.0227 0.0014 0.0007 0.0052** 0.1299 0.1935
(0.2063) (0.1404) (0.001) (0.4200) (0.1961) (0.1484) (0.0009) (0.07395) (0.0023) (0.0814) (0.1088)
e e e e 0.0414 0.0493 0.0211 0.0687 0.0142 0.1309** 0.0700 0.2873 0.0200** 0.0317 0.0900** 0.2605 0.0301 0.3845
(0.0443) (0.0350) (0.0482) (0.0494) (0.0147) (0.0600) (0.0531) (0.2011) (0.0096) (0.0224) (0.0349) (0.4079) (0.0271)
Cross-sectional
Individual FE
0.1405 0.0787 0.0008*** 0.8362 0.6528** 0.1416 0.0004 0.0140 0.0013 0.0059 0.0127 0.0167** 0.0906** 0.0043 0.0486 0.0025 0.0284 0.0116 0.0330 0.0077 0.0530 0.0211 0.03175 0.0042 0.0358 0.0346 0.0909 0.0519 0.1120
0.3814 0.0973 0.0002 1.5023*** 0.6242*** 0.3712** 0.0018 0.01715 0.0004 0.1917 0.1635 e e e e 0.0196 e 0.1977** 0.1325** 0.0132 0.0157 0.1413 0.2618 0.0027 0.0444 0.0797 0.8374 0.0249 0.0132
(0.1649) (0.0935) (0.0003) (0.5685) (0.2778) (0.1836) (0.0017) (0.0437) (0.0013) (0.0354) (0.0402) (0.0077) (0.0358) (0.0463) (0.0596) (0.0395) (0.1624) (0.0474) (0.0590) (0.0127) (0.0366) (0.0425) (0.0510) (0.0077) (0.0273) (0.0380) (0.2769) (0.0329)
(0.2099) (0.1648) (0.0004) (0.5435) (0.2079) (0.1695) (0.0015) (0.04385) (0.0014) (0.1026) (0.0868)
(0.0364) (0.0849) (0.0602) (0.0146) (0.0990) (0.1402) (0.2530) (0.0085) (0.0244) (0.0433) (0.5028) (0.0320)
Notes: All models included SNAP participation, two food price indices (fruit and vegetable and fast food meal), one food outlet density measure (the number of supermarkets/ grocery stores), and the interaction terms of SNAP participation with the food price and food outlet density variables as the main variables of interest. Match quality variable (distance from the centroid of zip code to the centroid of the nearest ACCRA city), age squared, and education missing indicator were all controlled for in the models but corresponding coefficients were not shown. Standard errors are in parentheses and were adjusted for clustering at the zip code level for cross-sectional and individual fixed effects models. **p 0.05, ***p 0.01.
E. Han et al. / Social Science & Medicine 74 (2012) 1874e1881
Simulations We conducted simulations to estimate whether the effects of policies related to economic contextual variables on BMI differed between participants and non-participants in SNAP. Our simulation results may be useful to approximate the potential effectiveness for SNAP versus non-SNAP participants of relevant policy measures aimed at reducing body mass such as subsidizing fruits and vegetables or improving access to supermarkets and grocery stores in underserved areas. The simulations revealed that for women a 20% reduction in the price of fruits and vegetables would decrease BMI more for SNAP participants than non-SNAP participants by 0.20 units in the crosssectional model, whereas it decreased BMI by only 0.026 units more for participants when individual permanent heterogeneity was controlled for. Increasing the number of supermarkets/grocery stores by 20% resulted in larger decreases in BMI among SNAP participants than non-participants by 0.032 and 0.031 units in the cross-sectional and individual fixed effects models, respectively. For men, the 20% simulated reduction of the price of fruits and vegetables yielded a 1.08 unit difference in BMI reduction between SNAP participants and non-participants. In the longitudinal individual fixed effects model, the difference of the reduction in BMI between SNAP participants and non-participants fell to 0.46 units. There was no statistically significant differential reduction in BMI between SNAP participants and non-participants with respect to a simulated increase in the number of supermarkets/grocery stores by 20% in either the cross-sectional or longitudinal model (Table 4). Discussion We explored the extent to which economic contextual factors moderated the association of SNAP participation with BMI among low-income adults whose per capita family income was less than 130% of the federal poverty guideline. Our study found a statistically significant joint moderating effect of the economic contextual factors in a longitudinal individual fixed effects model for both women (for BMI only) and men (for both BMI and obesity). We also found that the number of supermarkets/grocery stores had a statistically significant moderating effect on the rewarding effect of SNAP participation on BMI in the longitudinal individual fixed effects model for both women and men. A simulated policy change relevant to the economic contextual variables revealed a larger rewarding effect on BMI for SNAP participants compared to nonparticipants consequential to a simulated 20% decrease in the
Table 4 Simulated effects of policies related to economic contextual variables on BMI, by SNAP participation. Differences in BMI between participants and non-participants in SNAP Cross-sectional
Individual fixed effects
Women 20% Reduction in price of fruits 0.1969*** (0.0246) 0.0258*** (0.0042) and vegetables 20% Increase in # of supermarkets 0.0317*** (0.0419) 0.0305*** (0.4028) and grocery stores Men 20% Reduction in price of fruits 1.0826*** (0.1566) 0.4550*** (0.0617) and vegetables 20% Increase in # of supermarkets 0.0018 (0.0406) 0.0043 (0.0615) and grocery stores Notes: In all models, we controlled for the covariates as per the models estimated in Table 2. **p 0.05, ***p 0.01.
1879
price of fruits and vegetables (women and men) and to a 20% increase in the number of supermarkets and grocery stores (women only). A key strength of this study was that we combined data collected from multiple sources that allowed for the estimation of moderating effects of the community-level economic contextual factors in the association of SNAP participation with BMI at the individual-level. It is also a strength that we used longitudinal data to control for a potential source of the selection bias in individual SNAP participation, i.e. permanent individual heterogeneity. Given that time-constant unobserved individual heterogeneity is likely to affect individual SNAP participation decisions, individual food consumption patterns, and consequently weight outcomes, it is important to control for such sources of the bias. Most of the previous studies reported a significant positive association of SNAP participation with adult body weight outcomes only among women (Baum, 2007; Chen et al., 2005; Gibson, 2003; Townsend et al., 2001; Zagorsky & Smith, 2009). Studies using longitudinal analyses reported a substantial decrease in the magnitude of the estimated association between SNAP participation and BMI (Baum, 2007; Gibson, 2003; Kaushal, 2007). Similarly, we found that the association was statistically insignificant when we controlled for the time invariant endogeneity of SNAP participation using a longitudinal individual fixed effects model. We acknowledge a number of limitations of this study. The measurement of food prices based on ACCRA data is subject to several limitations. For example, the price data is based on establishment samples that reflect a higher standard of living in a limited number of larger cities and MSAs and the coverage of food items is limited (Powell & Bao, 2009b; Sturm & Datar, 2005). Nonetheless, given the national coverage of these price data, they have been similarly used in a number of previous studies (Chou et al., 2004; Chou, Rashad, & Grossman, 2008; Lakdawalla, Philipson, & Bhattacharya, 2005; Powell et al., 2011). For the density measures of food-related outlets, the accuracy of the D&B business list has been verified only in a small number of previous studies, and a wide variation in the extent of the accuracy has been reported (Kowaleski-Jones et al., 2009; Powell et al., 2011). However, the D&B list allows modeling such economic factors in a uniform manner in national sample. We also acknowledge that outlet densities may be endogenous. However, given that previous studies have showed food store availability to be associated with neighborhood income (Larson, Story, & Nelson, 2009; Powell, Chaloupka, & Bao, 2007), controlling for neighborhood median household income in our study might help to control for some of the neighborhood endogeniety. Although there is no food subsidy program that directly targets specific food items in the United States, California recently passed a legislation to conduct a “Healthy Purchase” pilot program where SNAP recipients will be subsidized a portion of the cost of purchasing fresh produce with food stamps (Guthrie, Frazão, Andrews, & Smallwood, 2007). Further, the Women, Infant and Children (WIC) program, a food assistance program in the United States, recently started offering additional monthly cash-value for fruits and vegetables in the amount of $10 for fully breastfeeding women, $8 for non-breastfeeding women, and $6 for children (Oliveira & Frazao, 2009). Also, Hampden County, Massachusetts, was chosen as the site for the Healthy Incentives Pilot (HIP) projects, which was funded under The Food, Nutrition and Conservation Act of 2008 to evaluate incentives provided to SNAP recipients at the point-of-sale to increase the purchase of fruits, vegetables or other healthful foods (Food and Nutrition Service, 2012). Our study findings from a simulated reduction in the price of fruits and vegetables by 20% that resulted in a larger decrease in
1880
E. Han et al. / Social Science & Medicine 74 (2012) 1874e1881
BMI among SNAP participants than non-participants for both women and men partly imply that subsidies for healthy food may be used as an effective tool to modify the relationship of SNAP participation on body mass outcomes. Further, we found that SNAP recipients’ BMI were more statistically significantly responsive to the availability of supermarkets/ grocery stores compared to non-recipients in a longitudinal model for both women and men. Similar findings have been reported in the previous literature. For example, Rose and Richards (2004) reported that food stamp recipients who lived closer to supermarkets consumed more fruits and vegetables (Rose & Richards, 2004). Studies also have reported that poor neighborhoods, where SNAP recipients are more likely to reside, had less access to supermarkets relative to affluent neighborhoods (Larson et al., 2009; Neckerman, Bader, Purciel, & Yousefzadeh, 2009; Neckerman, Lovasi, & Davies, 2009; Powell, Slater, Mirtcheva, Bao, & Chaloupka, 2007). Some studies also argued that poor neighborhoods may have more small groceries carrying various healthy food items (Bitler & Haider, 2009). Our results imply that increasing access to supermarkets/grocery stores which may carry more healthful items could be an effective tool to optimize the impact of SNAP participation on body mass. Fiscal policies to help SNAP participants to purchase more fruits and vegetables could reach its policy goal more effectively when combined with increased geographic accessibility to such healthy foods. At the same time, given that food stores are required to offer various and selected nutritious food products to participate in food assistance programs such as WIC (Oliveira & Frazao, 2009), the availability of healthy items in neighborhoods independent of store types may be more relevant in optimizing SNAP participation. Therefore, future research should further investigate the extent to which healthy foods are available within stores and its interrelationship with SNAP participation and food consumption and weight outcomes among people in low-socioeconomic status. Acknowledgment We gratefully acknowledge research support from the National Research Initiative of the U.S., Department of Agriculture Cooperative State Research, Education and Extension Service, grant number 2005-35215-15372. References Auld, M. C., & Powell, L. M. (2009). Economics of food energy density and adolescent body weight. Economica, 76(304), 719e740. Baum, C. (2007). The effects of food stamps on obesity. Washington, DC: United States Department of Agriculture Report No. 34. Besharov, D. (Dec 8 2002). We are feeding the poor as if they’re starving. Washington Post. Beydoun, M. A., Powell, L. M., & Wang, Y. (2008). Impacts of fast food, fruit and vegetable prices on dietary intakes among US adults: are they modified by family income? Social Science & Medicine, 66(11), 2218e2229. Bitler, M., & Haider, S. J. (2009). An economic view of food deserts in the United States. National Poverty Center Working Paper. Cawley, J. (2004). The impact of obesity on wages. Journal of Human Research, 39, 451e474. Chang, V. W., & Lauderdale, D. S. (2005). Income disparities in body mass index and obesity in the United States, 1971e2002. Archives of Internal Medicine, 165(18), 2122e2128. Chen, Z., Yen, S. T., & Eastwood, D. B. (2005). Effects of food stamp participation on body weight and obesity. American Journal of Agricultural Economics, 87(5), 1167e1173. Chou, S. Y., Grossman, M., & Saffer, H. (2004). An economic analysis of adult obesity: results from the behavioral risk factor surveillance system. Journal of Health Economics, 23(3), 565e587. Chou, S. Y., Rashad, I., & Grossman, M. (2008). Fast food restaurant advertising on television and its influence on childhood obesity. The Journal of Law and Economics, 51(4), 599e618. Dun and Bradstreet. (2005). The DUNS right quality process: The power behind quality information. Waltham, Mass: Dun and Bradstreet.
Finkelstein, E. A., Trogdon, J. G., Cohen, J. W., & Dietz, W. (2009). Annual medical spending attributable to obesity: payer-and service-specific estimates. Health Affairs, 28(5), 822e831. Flegal, K. M., Carroll, M. D., Ogden, C. L., & Curtin, L. R. (2010). Prevalence and trends in obesity among US adults, 1999e2008. JAMA: The Journal of the American Medical Association, 303(235), 235e241. Food and Nutrition Service. USDA. (2012). Healthy incentives pilot, supplemental nutrition assistance program. Downloaded from. http://www.fns.usda.gov/snap/ hip/default.htm. Fox, M. K., Hamilton, W., & Lin, B. H. (2004). Effects of food assistance and nutrition programs on healthIn Executive summary of the literature review, Vol. 4. Washington, DC: U.S. Dept. of Agriculture, Economic Research Service, Food Assistance and Nutrition Research Report No. 19e4. Fraker, T. M. (1990). The effects of food stamps on food consumption: A review of the literature. Washington, DC: U.S Department of Agriculture. Food and Consumer Services. Office of Analysis and Evaluation. Fraker, T. M., Martini, A. P., & Ohls, J. C. (1995). The effect of food stamp cashout on food expenditures: an assessment of the findings from four demonstrations. The Journal of Human Resources, 30(4), 633e649. Gibson, D. (2003). Food stamp program participation is positively related to obesity in low income women. Journal of Nutrition, 133(7), 2225e2231. Gibson, D. (2006). Long-term food stamp program participation is positively related to simultaneous overweight in young daughters and obesity in mothers. Journal of Nutrition, 136(4), 1081e1085. Guthrie, J., Frazão, E., Andrews, M., & Smallwood, D. (2007). Improving food choices e can food stamps do more? Amber Waves, 5(2), 22e28. Han, E., Norton, E. C., & Powell, L. M. (2011). Direct and indirect effects of body weight on adult wages. Economics & Human Biology, 9(4), 381e392. Han, E., Norton, E. C., & Stearns, S. C. (2009). Weight and wages: fat versus lean paychecks. Health Economics, 18, 535e548. Kaushal, N. (2007). Do food stamps cause obesity?: evidence from immigrant experience. Journal of Health Economics, 26(5), 968e991. Kowaleski-Jones, L., Fan, J. X., Yamada, I., Zick, C. D., Smith, K. R., & Brown, B. B. (2009). Alternative measures of food deserts: fruitful options or empty cupboards?. National Poverty Center Working Paper. Lakdawalla, D., Philipson, T., & Bhattacharya, J. (2005). Welfare-enhancing technological change and the growth of obesity. American Economic Review, 95(2), 253e257. Larson, N. I., Story, M. T., & Nelson, M. C. (2009). Neighborhood environments: disparities in access to healthy foods in the U.S. American Journal of Preventive Medicine, 36(1), 74e81. Meyerhoefer, C. D., & Pylypchuk, Y. (2008). Does participation in the food stamp program increase the prevalence of obesity and health care spending? American Journal of Agricultural Economics, 90(2), 287e305. Miljkovic, D., Nganje, W., & de Chastenet, H. (2008). Economic factors affecting the increase in obesity in the United States: differential response to price. Food Policy, 33(1), 48e60. Morland, K., Diez-Roux, A. V., & Wing, S. (2006). Supermarkets, other food stores, and obesity: the atherosclerosis risk in communities study. American Journal of Preventive Medicine, 30(4), 333e339. Morland, K., Wing, S., Diez-Roux, A., & Poole, C. (2002). Neighborhood characteristics associated with the location of food stores and food service places. American Journal of Preventive Medicine, 22(1), 23e29. Neckerman, K. M., Bader, M., Purciel, M., & Yousefzadeh, P. (2009). Measuring food access in urban areas. National Poverty Center Working Paper. Neckerman, K. M., Lovasi, G. S., & Davies, S. (2009). Disparities in urban neighborhood conditions: evidence from GIS measures and field observation in New York city. Journal of Public Health Policy, 30(S1), S264eS285. Office of Analysis. (2001). Making America stronger: A profile of the food stamp program 2005. Oliveira, V., & Frazao, E. (2009). The WIC program background, trends, and economic issues. U.S. Department of Agriculture, Economic Research Service. Economic Research Report No. 73. Powell, L. M., & Bao, Y. J. (2009). Food prices, access to food outlets and child weight outcomes. Economics & Human Biology, 7(1), 64e72. Powell, L. M., & Chaloupka, F. J. (2009). Food prices and obesity: evidence and policy implications for taxes and subsidies. The Milbank Quarterly, 87(1), 229e257. Powell, L. M., Chaloupka, F. J., & Bao, Y. (2007). The availability of fast-food and fullservice restaurants in the United States: associations with neighborhood characteristics. American Journal of Preventive Medicine, 33(4 Suppl), S240eS245. Powell, L. M., & Han, E. (2011). Adult obesity and the price and availability of food in the United States. American Journal of Agricultural Economics, 93(2), 378e384. Powell, L. M., Han, E., Khan, T., Quinn, C., Pugach, O., Gibbs, K., et al. (2011). Field validation of secondary commercial data sources on the retail food outlet environment in the U.S. Health & Place, 17(5), 1122e1131. Powell, L. M., Slater, S., Mirtcheva, D., Bao, Y., & Chaloupka, F. J. (2007). Food store availability and neighborhood characteristics in the United States. Preventive Medicine, 44(3), 189e195. Rhoades, J. A., Altman, B. M., & Cornelius, L. G. (2004). Trends in adult obesity in the United States, 1987 and 2001: estimates for the non-institutionalized population, age 20 to 64. Statistical brief no. 37. Rose, D., & Richards, R. (2004). Food store access and household fruit and vegetable use among participants in the US food stamp program. Public Health Nutrition, 7(8), 1081e1088.
E. Han et al. / Social Science & Medicine 74 (2012) 1874e1881 Stunkard, A. J., & Sorensen, T. I. (1993). Obesity and socioeconomic status e a complex relation. New England Journal of Medicine, 329(14), 1036e1037. Sturm, R., & Datar, A. (2005). Body mass index in elementary school children, metropolitan area food prices and food outlet density. Public Health, 119(12), 1059e1068. Townsend, M. S., Peerson, J., Love, B., Achterberg, C., & Murphy, S. P. (2001). Food insecurity is positively related to overweight in women. Journal of Nutrition, 131(6), 1738e1745. U.S. Department of Health and Human Services. (2001). The surgeon general’s call to action to prevent and decrease overweight and obesity. Rockville, MD: Public
1881
Health Service, Office of the Surgeon General. Accessed 27.09.10. http://www. surgeongeneral/gov/topics/obesity/calltoaction/toc.htm. US Department of Health and Human Services, National Center for Health Statistics. (1996). Third national health and nutrition examination survey, 1988e1994, NHANES III reference manuals and reports (CD-Rom). Hyattsville, MD: Centers for Disease Control and Prevention. Ver Ploeg, M., & Ralston, K. (2008). Food stamps and obesity: What do we know? U.S. Dept. of Agriculture, Economic Research Service. Report No. EIB-34. Zagorsky, J. L., & Smith, P. K. (2009). Does the U.S. food stamp program contribute to adult.
Contents lists available at SciVerse ScienceDirect
Social Science & Medicine journal homepage: www.elsevier.com/locate/socscimed
Supplemental nutrition assistance program and body weight outcomes: The role of economic contextual factors Euna Han a, *, Lisa M. Powell b, Zeynep Isgor c a
Gachon University, College of Pharmacy, 534-2 Yeonsu-Dong, Yeonsu-Gu, Inchon, 406-799, South Korea Department of Economics, Institute for Health Research and Policy, University of Illinois at Chicago, 1747 West Roosevelt Road, M/C 275, Chicago, IL 60608, USA c Department of Economics, University of Illinois at Chicago, 1747 West Roosevelt Road, M/C 275, Chicago, IL 60608, USA b
a r t i c l e i n f o
a b s t r a c t
Article history: Available online 20 March 2012
We explored the extent to which economic contextual factors moderated the association of Supplemental Nutrition Assistance Program (SNAP) participation with body mass index (BMI) among lowincome adults whose family income (adjusted for family size) is less than 130% of the federal poverty guideline. We drew on individual-level data from the Panel Study of Income Dynamics in the United States, including three waves of data in 1999, 2001, and 2003. Economic contextual data were drawn from the American Chamber of Commerce Researchers Association for food prices and Dun & Bradstreet for food outlet measures. In addition to cross-sectional estimation, a longitudinal individual fixed effects model was used to control for permanent unobserved individual heterogeneity. Our study found a statistically significant joint moderating effect of the economic contextual factors in longitudinal individual fixed effects model for both women (BMI only) and men (both BMI and obesity). For both women and men, SNAP participants’ BMI was statistically significantly lower if they faced increased numbers of available supermarkets/grocery stores in the longitudinal model. A simulated 20% reduction in the price of fruits and vegetables resulted in a larger decrease in BMI among SNAP participants than non-participants for women and men, whereas a simulated 20% increase in the availability of supermarkets and grocery stores resulted in a statistically significant difference in the change in BMI by SNAP participation for women but not for men. Policies related to economic contextual factors, such as subsidies for fruits and vegetables or those that would improve access to supermarkets and grocery stores may enhance the relationship between SNAP participation and body mass outcomes among food assistance program participants. Ó 2012 Elsevier Ltd. All rights reserved.
Keywords: Supplemental nutrition assistance program (SNAP) Body mass index (BMI) Economic contextual factors United States
Introduction The prevalence of obesity in the U.S. has risen at a striking rate over the past few decades, with one-third of American adults reported to be obese in 2007e08 (Flegal, Carroll, Ogden, & Curtin, 2010). Particularly, low-socioeconomic status (SES) individuals have shown higher prevalence of obesity (Chang & Lauderdale, 2005; Rhoades, Altman, & Cornelius, 2004; Stunkard & Sorensen, 1993; Townsend, Peerson, Love, Achterberg, & Murphy, 2001). Such phenomena have drawn a tremendous interest in understanding the contributors to this epidemic given the adverse associations of obesity with numerous health outcomes (USDHHS,
* Corresponding author. Tel.: þ82 32 820 4766; fax: þ82 32 820 4829. E-mail addresses: [email protected], [email protected] (E. Han), [email protected] (L.M. Powell), [email protected] (Z. Isgor). 0277-9536/$ e see front matter Ó 2012 Elsevier Ltd. All rights reserved. doi:10.1016/j.socscimed.2012.02.032
2001) and non-health outcomes such as wages (Han, Norton, Stearns, 2009; Han, Norton, & Powell, 2011), and the mounting societal burden from the high medical costs associated with obesity (Finkelstein, Trogdon, Cohen, & Dietz, 2009). Several previous studies have examined the Supplemental Nutrition Assistance Program (SNAP), previously called the Food Stamp Program, as one potential channel linked to high body weight outcomes for its eligible population (Baum, 2007; Chen, Yen, & Eastwood, 2005; Gibson, 2003, 2006; Meyerhoefer & Pylypchuk, 2008; Townsend et al., 2001; Zargosky & Smith, 2009). As the largest domestic food and nutrition assistance program in the U.S., the SNAP is one of the most important programs to combat food insecurity (Ver Ploeg & Ralston, 2008). Nonetheless, there have been ongoing debates on whether the provision of food vouchers has potentially encouraged its recipients to consume more food compared to cash assistance (Besharov, 2002; Fox, Hamilton, & Lin, 2004; Fraker, 1990; Fraker, Martini, &
E. Han et al. / Social Science & Medicine 74 (2012) 1874e1881
Ohls, 1995). Given that the SNAP benefits are entirely federally funded (Office of Analysis, 2001) and increased body weight puts substantial burden on the public (Finkelstein et al., 2009; USDHHS, 2001), it is important to explore the characteristics of the relationship of SNAP participation with body mass and effective policy measures to potentially complement the relationship so that the SNAP can improve participants’ body weight. An increasing number of studies have highlighted the importance of economic contextual factors such as food prices and food store outlet densities as important environmental factors that may contribute to the obesity epidemic. Higher fast food prices and lower grocery food prices (Chou, Grossman, & Saffer, 2004), higher prices of sugar (Miljkovic, Nganje, & de Chastenet, 2008), and lower fruit and vegetable prices (Beydoun, Powell, & Wang, 2008; Powell & Han, 2011) have been associated with lower body mass index (BMI) and obesity prevalence among adults. The geographic availability of foods through food stores also has been linked to the obesity epidemic. The greater availability of supermarkets has been associated with more fruit and vegetable intake, more healthful diets, and lower rates of obesity among American adults (Morland, Diez-Roux, & Wing, 2006; Morland, Wing, Diez-Roux, & Poole, 2002). This study built on the previous literature and examined the association of SNAP participation with body weight outcomes among people in low SES who are potential SNAP recipients, with a focus on whether economic contextual factors moderated the relationship between SNAP participation and body weight outcomes. Not only did we measure the association of the moderating effects of the economic contextual factors in a cross-sectional setting, we also estimated longitudinal individual-level fixed effects models using data from the Panel Study of Income Dynamics (PSID) in the United States to control for unobserved permanent individual heterogeneity. Methods Empirical models We regressed individual body weight outcomes (either linear BMI or a non-linear measure of obesity) on SNAP participation, economic contextual factors, and their interactions to measure the extent to which the economic contextual factors moderated the association of SNAP participation with body weight outcomes. We measured the economic contextual factors via the monetary cost of foods as well as opportunity cost of obtaining foods. The measurement of the monetary cost of foods included fruit and vegetable prices (proxy for healthy foods) and fast food prices (proxy for unhealthy foods). The opportunity cost of food consumption was measured by the geographic accessibility of food outlets including supermarkets and grocery stores. We controlled for neighborhood median household income as a proxy for neighborhood socioeconomic patterns. Our basic estimation was: SNAP
BMIist ¼ b0 þ b
SNAPPRICE
þb
þb
PRICE
SNAPist1 þ b
SNAPOC
PRICEst1 OC
SNAPist1 PRICEst1 þ b
OCst1
X
SNAPist1 OCst1 þ b Xist þ mi þ 3 ist (1)
where the subscripts i, t, and s denoted individual, year, and zip code, respectively. The b’s were parameters to be estimated. m was time invariant unobserved individual heterogeneity and 3 was a time-varying error term that was assumed to be normally distributed. Cross-sectional estimation does not control for the permanent portion of the error term, m, whereas the longitudinal
1875
individual fixed effects model controls for the time-constant unobserved individual heterogeneity. BMI was a linear measure of body mass calculated as weight in kilograms divided by height in meters squared. SNAP denoted SNAP participation. PRICE was a vector that measured the prices of fruits and vegetables and fast food in zip code s in year t 1. OC was a vector that included measures of the availability (per 10,000 capita per 10 square miles) of supermarkets and grocery stores in zip code s in year t 1. Xist was a vector of individual and household characteristics including year dummy variables. In the vector Xist, we also controlled for zip code level median household income and the price match quality measure of the distance in miles between the centroid of the zip code and the closest city in the price data. We reported the marginal effect of SNAP (bSNAP þ bSNAP PRICEPRICEst þ bSNAP OCOCst) which was calculated at the mean value of each economic contextual variable and all other covariates. We also reported the marginal effects of the economic contextual factors for both non-SNAP and SNAP recipients. The coefficients of the economic contextual variables (bPRICE and bOC) represent the marginal effect of a unit increase in those variables on BMI among non-SNAP participants. The marginal effect of a unit increase of each economic contextual variable on BMI among SNAP participants is the combined coefficients of an economic contextual variable and its interaction with SNAP (bPRICE þ bFSP PRICE and bOC þ bFSP OC). All the estimations were run separately by gender. In addition, we conducted an F-test to explore whether the moderating effects of the economic contextual variables were jointly statistically significant. We conducted simulations to assess whether the effects of policies related to the economic contextual variables on BMI differed between participants and non-participants in SNAP. For the simulations, we obtained differences in BMI before and after any simulated changes in economic contextual variables for participants and non-participants. We undertook simulations based on participation and non-participation and in each case reduced the price of fruits and vegetables by 20% or increased the number of supermarkets and grocery stores by 20%. Finally, we calculated the difference-in-difference as participants’ after and before simulated policy change difference minus the non-participants’ after and before simulated policy change difference. All standard errors were bootstrapped. This study was approved by the Institutional Review Board (IRB) of the University of Illinois at Chicago. Data Individual-level data We used the PSID in the United States, which began in 1968 as a longitudinal study of a representative sample of U.S. individuals and their co-residents. The PSID interviewed individuals from families in the core sample annually from 1968 to 1996, and biennially since 1997. The estimation sample for this study consisted of individuals whose per capita family income (adjusted for family size) was less than or equal to 130% of the federal poverty guideline. Weight status related measures became available in the PSID in 1999. Our estimation sample included three waves of data starting in 1999 and onwards (1999, 2001, and 2003). The final estimation samples of 2,391 women and 1,351 men were obtained after restricting the sample to those who were at or below 130% of the federal poverty line, between the ages of 18e65 years, not pregnant at the time of interview, and with non-missing information for any of the variables in the estimations.
1876
E. Han et al. / Social Science & Medicine 74 (2012) 1874e1881
Our dependent variable was BMI, measured as self-reported weight in kilograms divided by self-reported height in meters squared. Several previous studies, including Cawley (2004), attempted to adjust potential measurement errors for self-reported height and weight using the third National Health and Nutrition Examination Survey (NHANES III). However, we did not do so because the size or magnitude of the errors in self-reported height and weight in the NHANES III may be different than those in the PSID; in particular, the respondents in the NHANES III were aware that their weight and height would be measured after their selfreports of weight and height (Han, Norton, & Stearns, 2009; USDHHS, 1996). We further classified BMI into a clinical weight category of obese (BMI 30) as an additional dependent variable to investigate whether differential patterns of the moderating effects of economic contextual factors on the effect of SNAP participation on BMI were found at the right tail of the BMI distribution. SNAP participation, the primary independent variable of interest, was defined as a dichotomous indicator of whether the respondents or their spouses reported participating in the SNAP during the previous calendar year from the year of interview. Other covariates included the following: race/ethnicity (African American, Hispanic, other, with White as the reference), age, marital status (never married, separated, divorced, widowed, with married as the reference), number of children, education level (high school, some college, college or more, missing indicator for the education level, with less than high school as the reference), continuous measure of per capita family income, and year of the interview wave (2001 and 2003, with 1999 as the reference). We also controlled for the degree of urbanization of the individuals’ zip code of residence based on the Census 2000 data that measured population size within a zip code inside urbanized areas, outside urbanized areas (referred to as suburban areas), and in rural areas. Zip code level median household income based on the Census 2000 data was controlled for. Table 1 presents descriptive statistics of the estimation sample. The average BMI was 28.5 for women and 27.6 for men, and 35% and 26% of women and men, respectively, were classified as obese. Approximately 42% and 26% of women and men participated in SNAP during the previous calendar year, respectively. The average age of the sample persons was 38 years. Among women and men, respectively, approximately 29% and 39% were non-Hispanic white, 58% and 43% were non-Hispanic black, and 10% and 14% were Hispanic. Approximately one quarter of women (28%) and one half of men (47%) were married at the time of interview. Approximately one fourth of the sample population lived in a rural area for both genders. The approved amount for federal funding for the Food Stamp Nutrition Education Program (FSNE) per capita at state level was on average $0.3 in 1982e1984 dollars. Contextual program control We controlled for the approved amount of federal funding for the FSNE at the state level on a per capita basis as a proxy measure for unobserved state level effects of providing nutrition education to individuals that may increase knowledge and awareness regarding food intake. The FSNE is an important outreach program for low-income families and youth to adopt healthy lifestyles, such as consumption of more fruits, vegetables, whole grains, and lowfat dairy products, and engagement in daily physical activity. Between 1992 and 2004, approved federal funding that reimburses half of a state’s allowable costs increased from less than $1 million to $229 million (Office of Analysis, 2001). Economic contextual data Food price data were based on the American Chamber of Commerce Researchers Association (ACCRA) Cost of Living Index
Table 1 Summary statistics.
Dependent variables: BMI Obese Independent variables of interest SNAP participation Price of fruits and vegetables ($1982e84) Price of fast food ($1982e84) # of supermarkets and grocery storesb Average distance of price match (miles) Approved federal funding for Food Stamp Nutrition Education Program per capita state population ($1982e84) Age Race: Whitea Race: Black Race: Hispanic Race: Other race Marital status: Marrieda Marital status: Never married Marital status: Widowed Marital status: Divorced Marital status: Separated Number of children Education: Less than high schoola Education: High school Education: Some college Education: College or more Per capita family income ($1982e84) (`000) Median household income ($2000) (`000)c Urbanicity: Urbana Urbanicity: Suburban Urbanicity: Rural/Farm Year: 1999a Year: 2001 Year: 2003 N
Women
Men
All
All
28.4864 (7.2091) 0.3484
27.5756 (5.7316) 0.2591
0.4245 0.7783 (0.1062)
0.2554 0.7757(0.1124)
2.6941 (0.1595) 10.6383 (55.3120) 24.0131 (22.3081) 0.2973 (0.2778)
2.7177(0.1675) 8.7490 (27.9155) 26.6061 (24.1971) 0.3144 (0.2973)
37.6014 (11.3335) 0.2865 0.5826 0.1008 0.0301 0.2769 0.4040 0.0435 0.1706 0.1050 1.8076 (1.5464) 0.4262 0.3254 0.1547 0.0410 2.2980 (1.4200)
38.2258 (11.7433) 0.3901 0.4308 0.1429 0.0363 0.4730 0.3331 0.0126 0.1207 0.0607 1.3294 (1.5612) 0.4330 0.3175 0.1399 0.0725 2.5612 (1.6721)
33.2029 (10.6081) 35.0075 (12.0591) 0.6491 0.1397 0.2112 0.3446 0.3074 0.3480 2391
0.5892 0.1665 0.2443 0.3486 0.3020 0.3494 1351
Notes: Summary statistics are not weighted. Standard deviations (SD) are shown in parenthesis for continuous variables. a Denotes reference categories in regression models. Although the summary statistics for the missing indicator for the education variable and the age squared term are not included in the table, all regressions included these variables as control measures. b Food store related facility variable is defined per 10,000 capita per 10 square miles. c Median household income is based on Census 2000 Data and at the zip code level.
reports. These reports contain quarterly information on prices across more than 300 US cities and have been used in a number of previous studies (Auld & Powell, 2009; Chou et al., 2004; Powell et al., 2011; Powell & Chaloupka, 2009; Sturm & Datar, 2005). The price data were matched to the PSID sample based on the closest city match, which was determined by the shortest straight line distance between the centroid point of the adult’s zip code and the centroid point of the ACCRA price cities. We controlled for the price match quality using a variable based on this distance in miles in all estimations. We created the following two food price indices from the items provided in the ACCRA: 1) a fruit and vegetable price index as a weighted sum of the price of potatoes, bananas, lettuce, canned sweet peas, canned peaches, and frozen corn and 2) a fast food price index as a weighted sum of the prices of a McDonald’s QuarterPounder with cheese, a thin crust regular cheese pizza at Pizza Hut and/or Pizza Inn, and fried chicken (thigh and drumstick) at
E. Han et al. / Social Science & Medicine 74 (2012) 1874e1881
Kentucky Fried Chicken and/or Church’s Fried Chicken. Each of the price indices was weighted based on expenditure shares derived from the Bureau of Labor Statistics (BLS) Consumer Expenditure Survey. All prices were deflated by the BLS Consumer Price Index (1982e1984 ¼ 1). The average real price of fruits and vegetables and a fast food meal were $0.78 and $2.70, respectively, in our sample (Table 1). Data on food store outlets were drawn from a business list developed by Dun and Bradstreet (D&B) (Dun and Bradstreet, 2005). The outlet density data were matched to the PSID by year at the zip code level. We adjusted the number of available outlets per 10,000 capita per 10 square miles using the Census 2000 zip code level population and land area estimates in order to account for accessibility both in terms of congestion (per capita) and distance (per land area). We measured the number of supermarkets and grocery stores based on the 6-digit primary Standard Industrial Codes. There were on average 9e10 supermarkets/grocery stores per 10,000 capita per 10 squares miles per zip code with slight variations by gender (Table 1). All price and food outlet density measures were included in the estimation as one year lagged variables given that SNAP participation, the variable of interest, represented receipt of SNAP in the previous calendar year. Results BMI outcome The results indicated that female SNAP participants were estimated to have 1.31 units higher BMI than female non-participants
1877
when all covariates in the base model were set to their mean values in the cross-sectional model. We also rejected the null hypothesis of no joint moderating effects of the economic contextual variables (p-value ¼ 0.0122). An additional supermarket/grocery store in the area was estimated to decrease BMI by 0.0401 units for SNAP participants whereas for non-participants the corresponding estimate was 0.0029 and statistically insignificant. Controlling for permanent unobserved individual heterogeneity in the individual fixed effects model for women, we found no statistically significant relationship between SNAP participation and BMI. The economic contextual factors had joint moderating effects that increased the extent of the rewarding effect of SNAP participation on BMI among women given that the null hypothesis of no joint moderating effects was rejected (p-value ¼ 0.0147). Particularly, the rewarding effect of the number of supermarkets/ grocery stores remained statistically significant in the individual fixed effects model among SNAP participants (one additional supermarket/grocery store in the areas was associated with 0.0376 lower BMI units) whereas no such statistically significant effect was found for non-participants (see Table 2). For men, no statistically significant association of SNAP participation with BMI was found across all models. In the cross-sectional model, we found no statistically significant effects of economic contextual variables on BMI among either participants or nonparticipants in SNAP. In the individual fixed effects model controlling for time invariant unobserved individual heterogeneity, the moderating effect of the number of supermarkets/grocery stores was statistically significant for men. The number of supermarkets statistically significantly reduced BMI for both SNAP participants and non-
Table 2 Estimated regression coefficients from cross-sectional and longitudinal individual fixed effects models of BMI, by gender. Outcome variable: BMI
Price of fruits and vegetables Price of fast food # of supermarkets and grocery stores SNAP participation Price of fruits and vegetables SNAP participation Price of fast food SNAP participation # of supermarkets and grocery stores SNAP participation Funding for FSNE Program per capita state population ($1982e84) Median household income ($2000) (`000) Suburban Rural/farm Age Black Hispanic Other race Never married Widowed Divorced Separated Number of children Education: High school Education: Some college Education: College or more Per capita family income ($1982e84) (`000) Year: 2001 Year: 2003 Constant Marginal effect of SNAP participation P-value for joint F-test of all the moderating effects
Women (N ¼ 2391)
Men (N ¼ 1351)
Cross-sectional
Individual FE
Cross-sectional
Individual FE
2.1544 1.6637 0.0029 8.5028 1.3891 3.3896 0.0372*** 0.3023 0.0373** 0.2670 0.9741 0.1767 2.8132*** 0.8171 1.0624 1.0052** 0.8835 0.6073 0.2243 0.1687 0.2320 0.6656 0.4869 0.1165 0.0607 0.0901 25.042*** 1.3147*** 0.0122
2.1907 0.0786 0.0018 0.5536 0.1821 0.1894 0.0358*** 0.6845 0.0210 1.0915 0.9475 e e e e 0.4423 1.1392 0.4856 0.3192 0.3569** 0.3644 0.3433 2.3659 0.0471 0.5311** 0.7483** 24.466*** 0.2823 0.0147
0.6808 0.8005 0.0050 9.0352 7.0760 1.5535 0.0090 0.6054 0.0133 0.3099 0.2035 0.2994*** 0.8728 0.0001 0.7547 0.1838 1.0061 0.2782 0.1658 0.1841 0.8881 0.2900 0.3433 0.1049 0.4076 0.4263 21.007*** 0.5972 0.3490
2.4910 2.1917 0.0114** 16.052*** 2.9742 5.1559*** 0.0213 0.3071 0.0225 0.8452 0.4648 e e e e 0.9704 e 0.8053 1.6696 0.1510 0.1523 2.6838** 1.5868 0.1838** 0.6199** 1.0798** 35.199*** 0.0815 0.0312
(2.2222) (1.4203) (0.0016) (5.5008) (2.7434) (1.9035) (0.0129) (0.7085) (0.0185) (0.6588) (0.5444) (0.1100) (0.4725) (0.7544) (0.8782) (0.4901) (0.9635) (0.5704) (0.6836) (0.1423) (0.4449) (0.5164) (0.7544) (0.1144) (0.3420) (0.4516) (4.2949) (0.4073)
(1.8434) (1.1671) (0.0047) (4.2141) (1.9192) (1.4270) (0.0110) (0.6464) (0.0170) (0.5785) (0.6147)
(0.4316) (1.0083) (0.5714) (0.5583) (0.1583) (0.6601) (0.6578) (1.4757) (0.0860) (0.2221) (0.3381) (3.3388) (0.3065)
(2.5856) (1.2926) (0.0037) (7.2030) (4.0452) (2.2840) (0.0179) (0.6349) (0.0162) (0.4754) (0.5099) (0.0974) (0.4816) (0.6461) (0.6982) (0.4706) (2.1504) (0.6880) (0.9750) (0.1519) (0.4806) (0.5100) (0.7039) (0.0890) (0.3899) (0.5136) (4.2988) (0.4494)
(3.4805) (1.9478) (0.0050) (5.9614) (2.8090) (1.9613) (0.0154) (0.4978) (0.0151) (1.0192) (0.8162)
(0.6713) (0.8149) (0.9745) (0.2388) (0.8009) (1.1917) (1.6377) (0.0896) (0.2835) (0.4611) (6.8372) (0.3234)
Notes: All models included SNAP participation, two food price indices (fruit and vegetable and fast food meal), one food outlet density measure (the number of supermarkets/ grocery stores), and the interaction terms of SNAP participation with the food price and food outlet density variables as the main variables of interest. Match quality variable (distance from the centroid of zip code to the centroid of the nearest ACCRA city), age squared, and education missing indicator were all controlled for in the models but corresponding coefficients were not shown. Standard errors are in parentheses and were adjusted for clustering at the zip code level for cross-sectional and individual fixed effects models. **p 0.05, ***p 0.01.
1878
E. Han et al. / Social Science & Medicine 74 (2012) 1874e1881
participants with a bigger effect for participants. Specifically, one additional supermarket/grocery store was estimated to decrease BMI by 0.0114 units among non-participants in SNAP whereas the size of the decrease was estimated to be 0.0326 units for participants. Further, the null hypothesis of no joint moderating effects of the economic contextual variables was rejected (p-value ¼ 0.0312) (see Table 2). For other covariates included in all models, the signs of the coefficients were as we expected in general. Non-Hispanic black women had higher BMI by 2.8 unites than non-Hispanic white women. Women with more children had higher BMI when unobserved individual heterogeneity was controlled for in the longitudinal model. There was an upward time trend of average BMI for both genders to the extent that the average BMI increased by approximately one unit in year 2003 compared to year 1999.
would decrease the likelihood of obesity by 0.03 and 0.19 percentage points for non-participants and participants in SNAP, respectively, in the cross-sectional model. The effect of the number of supermarkets/grocery stores on the likelihood of obesity turned statistically insignificant in the longitudinal individual fixed effects model (Table 3). For men, one more supermarket/grocery store decreased the likelihood of obesity by 0.08 percentage points for only nonparticipants in SNAP in the cross-sectional model. Further controlling for the permanent individual heterogeneity in the longitudinal individual fixed effects model revealed no statistically significant effects of the economic contextual variables regardless of SNAP participation. The results also showed that the null hypothesis of no joint moderating effect was rejected in the individual fixed effects model (p-value ¼ 0.0132). The marginal effect of SNAP participation on the likelihood of obesity was negative but statistically insignificant.
Obesity outcome SNAP participant women showed higher likelihood of being obese by 7.3 percentage points in the cross-sectional model when all other covariates were set to their mean values. The economic contextual factors jointly reduced the penalizing effect of SNAP participation on the likelihood of obesity in the cross-sectional model for women (p-value ¼ 0.0073 for the null hypothesis of no joint moderating effects). The joint moderating effect was no longer statistically significant when time invariant individual heterogeneity was further controlled for in the individual fixed effects model. We also found a small negative (rewarding) effect of the number of supermarkets/grocery stores for the likelihood of obesity to the extent that one more supermarket/grocery store
Sensitivity analyses We performed sensitivity analyses by replicating all estimations including only the price variables as economic contextual measures to ensure our main findings. This was to test whether the estimation results with only food price variables were comparable to when the arguably endogenous outlet density variables were included. The estimates for SNAP participation, food prices, and the interaction terms of SNAP participation and food prices overall remained similar in terms of direction, magnitude, and statistical significance between those two specifications (results not shown in Tables).
Table 3 Estimated regression coefficients from cross-sectional and longitudinal individual fixed effects models of the probability of obesity, by gender. Outcome variable: OBESE
Women (N ¼ 2391) Cross-sectional
Price of fruits and vegetables Price of fast food # of supermarkets and grocery stores SNAP participation Price of fruits and vegetables SNAP participation Price of fast food SNAP participation # of supermarkets and grocery stores SNAP participation Funding for FSNE Program per capita state population ($1982e84) Median household income ($2000) (`000) Suburban Rural/farm Age Black Hispanic Other race Never married Widowed Divorced Separated Number of children Education: High school Education: Some college Education: College or more Per capita family income ($1982e84) (`000) Year: 2001 Year: 2003 Constant Marginal effect of SNAP participation P-value for joint F-test of all the moderating effects
0.0741 0.1122 0.0003*** 0.8080** 0.0523 0.3184** 0.0017*** 0.0437 0.0016 0.0739 0.1015*** 0.0056 0.1688*** 0.0067 0.0627 0.0625 0.0972 0.0327 0.0493 0.0099 0.0122 0.0048 0.0249 0.0054 0.0059 0.0050 0.3171 0.0728*** 0.0073
(0.1561) (0.1019) (0.0001) (0.3922) (0.2035) (0.1365) (0.0006) (0.0521) (0.0011) (0.0444) (0.0377) (0.0068) (0.0328) (0.0479) (0.0598) (0.0329) (0.0588) (0.0347) (0.0426) (0.0091) (0.0275) (0.0351) (0.0509) (0.0080) (0.0232) (0.0313) (0.2951) (0.0256)
Men (N ¼ 1351) Individual FE 0.0192 0.0217 0.0004 0.0700 0.0302 0.0227 0.0014 0.0007 0.0052** 0.1299 0.1935
(0.2063) (0.1404) (0.001) (0.4200) (0.1961) (0.1484) (0.0009) (0.07395) (0.0023) (0.0814) (0.1088)
e e e e 0.0414 0.0493 0.0211 0.0687 0.0142 0.1309** 0.0700 0.2873 0.0200** 0.0317 0.0900** 0.2605 0.0301 0.3845
(0.0443) (0.0350) (0.0482) (0.0494) (0.0147) (0.0600) (0.0531) (0.2011) (0.0096) (0.0224) (0.0349) (0.4079) (0.0271)
Cross-sectional
Individual FE
0.1405 0.0787 0.0008*** 0.8362 0.6528** 0.1416 0.0004 0.0140 0.0013 0.0059 0.0127 0.0167** 0.0906** 0.0043 0.0486 0.0025 0.0284 0.0116 0.0330 0.0077 0.0530 0.0211 0.03175 0.0042 0.0358 0.0346 0.0909 0.0519 0.1120
0.3814 0.0973 0.0002 1.5023*** 0.6242*** 0.3712** 0.0018 0.01715 0.0004 0.1917 0.1635 e e e e 0.0196 e 0.1977** 0.1325** 0.0132 0.0157 0.1413 0.2618 0.0027 0.0444 0.0797 0.8374 0.0249 0.0132
(0.1649) (0.0935) (0.0003) (0.5685) (0.2778) (0.1836) (0.0017) (0.0437) (0.0013) (0.0354) (0.0402) (0.0077) (0.0358) (0.0463) (0.0596) (0.0395) (0.1624) (0.0474) (0.0590) (0.0127) (0.0366) (0.0425) (0.0510) (0.0077) (0.0273) (0.0380) (0.2769) (0.0329)
(0.2099) (0.1648) (0.0004) (0.5435) (0.2079) (0.1695) (0.0015) (0.04385) (0.0014) (0.1026) (0.0868)
(0.0364) (0.0849) (0.0602) (0.0146) (0.0990) (0.1402) (0.2530) (0.0085) (0.0244) (0.0433) (0.5028) (0.0320)
Notes: All models included SNAP participation, two food price indices (fruit and vegetable and fast food meal), one food outlet density measure (the number of supermarkets/ grocery stores), and the interaction terms of SNAP participation with the food price and food outlet density variables as the main variables of interest. Match quality variable (distance from the centroid of zip code to the centroid of the nearest ACCRA city), age squared, and education missing indicator were all controlled for in the models but corresponding coefficients were not shown. Standard errors are in parentheses and were adjusted for clustering at the zip code level for cross-sectional and individual fixed effects models. **p 0.05, ***p 0.01.
E. Han et al. / Social Science & Medicine 74 (2012) 1874e1881
Simulations We conducted simulations to estimate whether the effects of policies related to economic contextual variables on BMI differed between participants and non-participants in SNAP. Our simulation results may be useful to approximate the potential effectiveness for SNAP versus non-SNAP participants of relevant policy measures aimed at reducing body mass such as subsidizing fruits and vegetables or improving access to supermarkets and grocery stores in underserved areas. The simulations revealed that for women a 20% reduction in the price of fruits and vegetables would decrease BMI more for SNAP participants than non-SNAP participants by 0.20 units in the crosssectional model, whereas it decreased BMI by only 0.026 units more for participants when individual permanent heterogeneity was controlled for. Increasing the number of supermarkets/grocery stores by 20% resulted in larger decreases in BMI among SNAP participants than non-participants by 0.032 and 0.031 units in the cross-sectional and individual fixed effects models, respectively. For men, the 20% simulated reduction of the price of fruits and vegetables yielded a 1.08 unit difference in BMI reduction between SNAP participants and non-participants. In the longitudinal individual fixed effects model, the difference of the reduction in BMI between SNAP participants and non-participants fell to 0.46 units. There was no statistically significant differential reduction in BMI between SNAP participants and non-participants with respect to a simulated increase in the number of supermarkets/grocery stores by 20% in either the cross-sectional or longitudinal model (Table 4). Discussion We explored the extent to which economic contextual factors moderated the association of SNAP participation with BMI among low-income adults whose per capita family income was less than 130% of the federal poverty guideline. Our study found a statistically significant joint moderating effect of the economic contextual factors in a longitudinal individual fixed effects model for both women (for BMI only) and men (for both BMI and obesity). We also found that the number of supermarkets/grocery stores had a statistically significant moderating effect on the rewarding effect of SNAP participation on BMI in the longitudinal individual fixed effects model for both women and men. A simulated policy change relevant to the economic contextual variables revealed a larger rewarding effect on BMI for SNAP participants compared to nonparticipants consequential to a simulated 20% decrease in the
Table 4 Simulated effects of policies related to economic contextual variables on BMI, by SNAP participation. Differences in BMI between participants and non-participants in SNAP Cross-sectional
Individual fixed effects
Women 20% Reduction in price of fruits 0.1969*** (0.0246) 0.0258*** (0.0042) and vegetables 20% Increase in # of supermarkets 0.0317*** (0.0419) 0.0305*** (0.4028) and grocery stores Men 20% Reduction in price of fruits 1.0826*** (0.1566) 0.4550*** (0.0617) and vegetables 20% Increase in # of supermarkets 0.0018 (0.0406) 0.0043 (0.0615) and grocery stores Notes: In all models, we controlled for the covariates as per the models estimated in Table 2. **p 0.05, ***p 0.01.
1879
price of fruits and vegetables (women and men) and to a 20% increase in the number of supermarkets and grocery stores (women only). A key strength of this study was that we combined data collected from multiple sources that allowed for the estimation of moderating effects of the community-level economic contextual factors in the association of SNAP participation with BMI at the individual-level. It is also a strength that we used longitudinal data to control for a potential source of the selection bias in individual SNAP participation, i.e. permanent individual heterogeneity. Given that time-constant unobserved individual heterogeneity is likely to affect individual SNAP participation decisions, individual food consumption patterns, and consequently weight outcomes, it is important to control for such sources of the bias. Most of the previous studies reported a significant positive association of SNAP participation with adult body weight outcomes only among women (Baum, 2007; Chen et al., 2005; Gibson, 2003; Townsend et al., 2001; Zagorsky & Smith, 2009). Studies using longitudinal analyses reported a substantial decrease in the magnitude of the estimated association between SNAP participation and BMI (Baum, 2007; Gibson, 2003; Kaushal, 2007). Similarly, we found that the association was statistically insignificant when we controlled for the time invariant endogeneity of SNAP participation using a longitudinal individual fixed effects model. We acknowledge a number of limitations of this study. The measurement of food prices based on ACCRA data is subject to several limitations. For example, the price data is based on establishment samples that reflect a higher standard of living in a limited number of larger cities and MSAs and the coverage of food items is limited (Powell & Bao, 2009b; Sturm & Datar, 2005). Nonetheless, given the national coverage of these price data, they have been similarly used in a number of previous studies (Chou et al., 2004; Chou, Rashad, & Grossman, 2008; Lakdawalla, Philipson, & Bhattacharya, 2005; Powell et al., 2011). For the density measures of food-related outlets, the accuracy of the D&B business list has been verified only in a small number of previous studies, and a wide variation in the extent of the accuracy has been reported (Kowaleski-Jones et al., 2009; Powell et al., 2011). However, the D&B list allows modeling such economic factors in a uniform manner in national sample. We also acknowledge that outlet densities may be endogenous. However, given that previous studies have showed food store availability to be associated with neighborhood income (Larson, Story, & Nelson, 2009; Powell, Chaloupka, & Bao, 2007), controlling for neighborhood median household income in our study might help to control for some of the neighborhood endogeniety. Although there is no food subsidy program that directly targets specific food items in the United States, California recently passed a legislation to conduct a “Healthy Purchase” pilot program where SNAP recipients will be subsidized a portion of the cost of purchasing fresh produce with food stamps (Guthrie, Frazão, Andrews, & Smallwood, 2007). Further, the Women, Infant and Children (WIC) program, a food assistance program in the United States, recently started offering additional monthly cash-value for fruits and vegetables in the amount of $10 for fully breastfeeding women, $8 for non-breastfeeding women, and $6 for children (Oliveira & Frazao, 2009). Also, Hampden County, Massachusetts, was chosen as the site for the Healthy Incentives Pilot (HIP) projects, which was funded under The Food, Nutrition and Conservation Act of 2008 to evaluate incentives provided to SNAP recipients at the point-of-sale to increase the purchase of fruits, vegetables or other healthful foods (Food and Nutrition Service, 2012). Our study findings from a simulated reduction in the price of fruits and vegetables by 20% that resulted in a larger decrease in
1880
E. Han et al. / Social Science & Medicine 74 (2012) 1874e1881
BMI among SNAP participants than non-participants for both women and men partly imply that subsidies for healthy food may be used as an effective tool to modify the relationship of SNAP participation on body mass outcomes. Further, we found that SNAP recipients’ BMI were more statistically significantly responsive to the availability of supermarkets/ grocery stores compared to non-recipients in a longitudinal model for both women and men. Similar findings have been reported in the previous literature. For example, Rose and Richards (2004) reported that food stamp recipients who lived closer to supermarkets consumed more fruits and vegetables (Rose & Richards, 2004). Studies also have reported that poor neighborhoods, where SNAP recipients are more likely to reside, had less access to supermarkets relative to affluent neighborhoods (Larson et al., 2009; Neckerman, Bader, Purciel, & Yousefzadeh, 2009; Neckerman, Lovasi, & Davies, 2009; Powell, Slater, Mirtcheva, Bao, & Chaloupka, 2007). Some studies also argued that poor neighborhoods may have more small groceries carrying various healthy food items (Bitler & Haider, 2009). Our results imply that increasing access to supermarkets/grocery stores which may carry more healthful items could be an effective tool to optimize the impact of SNAP participation on body mass. Fiscal policies to help SNAP participants to purchase more fruits and vegetables could reach its policy goal more effectively when combined with increased geographic accessibility to such healthy foods. At the same time, given that food stores are required to offer various and selected nutritious food products to participate in food assistance programs such as WIC (Oliveira & Frazao, 2009), the availability of healthy items in neighborhoods independent of store types may be more relevant in optimizing SNAP participation. Therefore, future research should further investigate the extent to which healthy foods are available within stores and its interrelationship with SNAP participation and food consumption and weight outcomes among people in low-socioeconomic status. Acknowledgment We gratefully acknowledge research support from the National Research Initiative of the U.S., Department of Agriculture Cooperative State Research, Education and Extension Service, grant number 2005-35215-15372. References Auld, M. C., & Powell, L. M. (2009). Economics of food energy density and adolescent body weight. Economica, 76(304), 719e740. Baum, C. (2007). The effects of food stamps on obesity. Washington, DC: United States Department of Agriculture Report No. 34. Besharov, D. (Dec 8 2002). We are feeding the poor as if they’re starving. Washington Post. Beydoun, M. A., Powell, L. M., & Wang, Y. (2008). Impacts of fast food, fruit and vegetable prices on dietary intakes among US adults: are they modified by family income? Social Science & Medicine, 66(11), 2218e2229. Bitler, M., & Haider, S. J. (2009). An economic view of food deserts in the United States. National Poverty Center Working Paper. Cawley, J. (2004). The impact of obesity on wages. Journal of Human Research, 39, 451e474. Chang, V. W., & Lauderdale, D. S. (2005). Income disparities in body mass index and obesity in the United States, 1971e2002. Archives of Internal Medicine, 165(18), 2122e2128. Chen, Z., Yen, S. T., & Eastwood, D. B. (2005). Effects of food stamp participation on body weight and obesity. American Journal of Agricultural Economics, 87(5), 1167e1173. Chou, S. Y., Grossman, M., & Saffer, H. (2004). An economic analysis of adult obesity: results from the behavioral risk factor surveillance system. Journal of Health Economics, 23(3), 565e587. Chou, S. Y., Rashad, I., & Grossman, M. (2008). Fast food restaurant advertising on television and its influence on childhood obesity. The Journal of Law and Economics, 51(4), 599e618. Dun and Bradstreet. (2005). The DUNS right quality process: The power behind quality information. Waltham, Mass: Dun and Bradstreet.
Finkelstein, E. A., Trogdon, J. G., Cohen, J. W., & Dietz, W. (2009). Annual medical spending attributable to obesity: payer-and service-specific estimates. Health Affairs, 28(5), 822e831. Flegal, K. M., Carroll, M. D., Ogden, C. L., & Curtin, L. R. (2010). Prevalence and trends in obesity among US adults, 1999e2008. JAMA: The Journal of the American Medical Association, 303(235), 235e241. Food and Nutrition Service. USDA. (2012). Healthy incentives pilot, supplemental nutrition assistance program. Downloaded from. http://www.fns.usda.gov/snap/ hip/default.htm. Fox, M. K., Hamilton, W., & Lin, B. H. (2004). Effects of food assistance and nutrition programs on healthIn Executive summary of the literature review, Vol. 4. Washington, DC: U.S. Dept. of Agriculture, Economic Research Service, Food Assistance and Nutrition Research Report No. 19e4. Fraker, T. M. (1990). The effects of food stamps on food consumption: A review of the literature. Washington, DC: U.S Department of Agriculture. Food and Consumer Services. Office of Analysis and Evaluation. Fraker, T. M., Martini, A. P., & Ohls, J. C. (1995). The effect of food stamp cashout on food expenditures: an assessment of the findings from four demonstrations. The Journal of Human Resources, 30(4), 633e649. Gibson, D. (2003). Food stamp program participation is positively related to obesity in low income women. Journal of Nutrition, 133(7), 2225e2231. Gibson, D. (2006). Long-term food stamp program participation is positively related to simultaneous overweight in young daughters and obesity in mothers. Journal of Nutrition, 136(4), 1081e1085. Guthrie, J., Frazão, E., Andrews, M., & Smallwood, D. (2007). Improving food choices e can food stamps do more? Amber Waves, 5(2), 22e28. Han, E., Norton, E. C., & Powell, L. M. (2011). Direct and indirect effects of body weight on adult wages. Economics & Human Biology, 9(4), 381e392. Han, E., Norton, E. C., & Stearns, S. C. (2009). Weight and wages: fat versus lean paychecks. Health Economics, 18, 535e548. Kaushal, N. (2007). Do food stamps cause obesity?: evidence from immigrant experience. Journal of Health Economics, 26(5), 968e991. Kowaleski-Jones, L., Fan, J. X., Yamada, I., Zick, C. D., Smith, K. R., & Brown, B. B. (2009). Alternative measures of food deserts: fruitful options or empty cupboards?. National Poverty Center Working Paper. Lakdawalla, D., Philipson, T., & Bhattacharya, J. (2005). Welfare-enhancing technological change and the growth of obesity. American Economic Review, 95(2), 253e257. Larson, N. I., Story, M. T., & Nelson, M. C. (2009). Neighborhood environments: disparities in access to healthy foods in the U.S. American Journal of Preventive Medicine, 36(1), 74e81. Meyerhoefer, C. D., & Pylypchuk, Y. (2008). Does participation in the food stamp program increase the prevalence of obesity and health care spending? American Journal of Agricultural Economics, 90(2), 287e305. Miljkovic, D., Nganje, W., & de Chastenet, H. (2008). Economic factors affecting the increase in obesity in the United States: differential response to price. Food Policy, 33(1), 48e60. Morland, K., Diez-Roux, A. V., & Wing, S. (2006). Supermarkets, other food stores, and obesity: the atherosclerosis risk in communities study. American Journal of Preventive Medicine, 30(4), 333e339. Morland, K., Wing, S., Diez-Roux, A., & Poole, C. (2002). Neighborhood characteristics associated with the location of food stores and food service places. American Journal of Preventive Medicine, 22(1), 23e29. Neckerman, K. M., Bader, M., Purciel, M., & Yousefzadeh, P. (2009). Measuring food access in urban areas. National Poverty Center Working Paper. Neckerman, K. M., Lovasi, G. S., & Davies, S. (2009). Disparities in urban neighborhood conditions: evidence from GIS measures and field observation in New York city. Journal of Public Health Policy, 30(S1), S264eS285. Office of Analysis. (2001). Making America stronger: A profile of the food stamp program 2005. Oliveira, V., & Frazao, E. (2009). The WIC program background, trends, and economic issues. U.S. Department of Agriculture, Economic Research Service. Economic Research Report No. 73. Powell, L. M., & Bao, Y. J. (2009). Food prices, access to food outlets and child weight outcomes. Economics & Human Biology, 7(1), 64e72. Powell, L. M., & Chaloupka, F. J. (2009). Food prices and obesity: evidence and policy implications for taxes and subsidies. The Milbank Quarterly, 87(1), 229e257. Powell, L. M., Chaloupka, F. J., & Bao, Y. (2007). The availability of fast-food and fullservice restaurants in the United States: associations with neighborhood characteristics. American Journal of Preventive Medicine, 33(4 Suppl), S240eS245. Powell, L. M., & Han, E. (2011). Adult obesity and the price and availability of food in the United States. American Journal of Agricultural Economics, 93(2), 378e384. Powell, L. M., Han, E., Khan, T., Quinn, C., Pugach, O., Gibbs, K., et al. (2011). Field validation of secondary commercial data sources on the retail food outlet environment in the U.S. Health & Place, 17(5), 1122e1131. Powell, L. M., Slater, S., Mirtcheva, D., Bao, Y., & Chaloupka, F. J. (2007). Food store availability and neighborhood characteristics in the United States. Preventive Medicine, 44(3), 189e195. Rhoades, J. A., Altman, B. M., & Cornelius, L. G. (2004). Trends in adult obesity in the United States, 1987 and 2001: estimates for the non-institutionalized population, age 20 to 64. Statistical brief no. 37. Rose, D., & Richards, R. (2004). Food store access and household fruit and vegetable use among participants in the US food stamp program. Public Health Nutrition, 7(8), 1081e1088.
E. Han et al. / Social Science & Medicine 74 (2012) 1874e1881 Stunkard, A. J., & Sorensen, T. I. (1993). Obesity and socioeconomic status e a complex relation. New England Journal of Medicine, 329(14), 1036e1037. Sturm, R., & Datar, A. (2005). Body mass index in elementary school children, metropolitan area food prices and food outlet density. Public Health, 119(12), 1059e1068. Townsend, M. S., Peerson, J., Love, B., Achterberg, C., & Murphy, S. P. (2001). Food insecurity is positively related to overweight in women. Journal of Nutrition, 131(6), 1738e1745. U.S. Department of Health and Human Services. (2001). The surgeon general’s call to action to prevent and decrease overweight and obesity. Rockville, MD: Public
1881
Health Service, Office of the Surgeon General. Accessed 27.09.10. http://www. surgeongeneral/gov/topics/obesity/calltoaction/toc.htm. US Department of Health and Human Services, National Center for Health Statistics. (1996). Third national health and nutrition examination survey, 1988e1994, NHANES III reference manuals and reports (CD-Rom). Hyattsville, MD: Centers for Disease Control and Prevention. Ver Ploeg, M., & Ralston, K. (2008). Food stamps and obesity: What do we know? U.S. Dept. of Agriculture, Economic Research Service. Report No. EIB-34. Zagorsky, J. L., & Smith, P. K. (2009). Does the U.S. food stamp program contribute to adult.