Original Article
The Supplemental Nutrition Assistance Program and frequency of sugar-sweetened soft drink consumption among low-income adults in the US
Nutrition and Health 1–11 ª The Author(s) 2017 Reprints and permission: sagepub.co.uk/journalsPermissions.nav DOI: 10.1177/0260106017726248 journals.sagepub.com/home/nah
Jiyeun Park1, Hsien-Chang Lin2 and Chao-Ying Peng3
Abstract Background: The Supplemental Nutrition Assistance Program (SNAP) was designed to help low-income people purchase nutritious foods in the US. In recent years, there has been a consistent call for banning purchases of sugar drinks in SNAP. Aim: The aim of this study was to examine the association between SNAP participation and the frequency of sugar-sweetened soft drink (SSD) consumption among low-income adults in the US. Method: Data came from the 2009– 2010 National Health and Nutrition Examination Survey. Low-income adults aged 20 years with a household income 250% of the Federal Poverty Level (N ¼ 1200) were categorized into two groups based on the household’s SNAP receipt: SNAP recipients (n ¼ 393) and non-recipients (n ¼ 807). Propensity-score matching was used to minimize observable differences between these two groups that may explain the difference in SSD consumption, generating the final sample of 393 matched pairs (SNAP recipients, n ¼ 393; non-recipients, n ¼ 393). An ordinal logistic regression was conducted on the matched sample. Results: SNAP recipients were more likely to report higher levels of SSD consumption, compared with non-recipients (adjusted odds ratio (AOR) ¼ 1.55, 95% confidence interval (CI) ¼ 1.172.07). Male gender (AOR ¼ 1.69, 95% CI ¼ 1.172.46), younger age (AOR ¼ 0.97, 95% CI ¼ 0.960.99), lower education level (AOR ¼ 2.28, 95% CI ¼ 1.333.89), and soda availability in homes (AOR ¼ 2.24, 95% CI ¼ 1.772.83) were also associated with higher levels of SSD consumption among low-income adults. Conclusions: SNAP participation was associated with frequent SSD consumption. To reduce SSD consumption, strategic efforts need to focus on educating people about the harms of SSD and promoting nutritious food choices with SNAP benefits. Keywords Sugar-sweetened soft drink, Supplemental Nutrition Assistance Program, low-income, policy, nutrition
Introduction The Supplemental Nutrition Assistance Program (SNAP), formerly known as the Food Stamp Program (FSP), is the largest nutrition assistance program administered by the US Department of Agriculture (USDA). During the fiscal year 2012, more than 46 million people participated in the program in an average month, leading to over $74 billion SNAP spending in 2012 (Cunnyngham, 2012). SNAP was designed to alleviate hunger and malnutrition among lowincome households by increasing their food purchasing power for nutritious foods (Cunnyngham, 2012). Lowincome households with a gross monthly income 130% of Federal Poverty Level (FPL) are eligible to receive SNAP benefits that can be used to purchase most food items such as breads, cereal, fruits and vegetables, meats, fish and poultry, dairy products, and sugar drinks. Alcoholic beverages, tobacco products, non-food items, such as
medicine and vitamins, and household supplies are ineligible for the purchases with SNAP benefits. The extent to which FSP/SNAP was associated with food insecurity or diet quality has been studied extensively by many researchers; FSP was associated with more food expenditures and improved diet quality (Basiotis et al.,
1
Department of Psychiatry and Behavioral Sciences, School of Medicine, Duke University, Durham, NC, USA 2 Department of Applied Health Science, School of Public Health, Indiana University, Bloomington, IN, USA 3 School of Education, Indiana University, Bloomington, IN, USA Corresponding author: Hsien-Chang Lin, Department of Applied Health Science, School of Public Health, Indiana University, 1025 East 7th Street, SPH 116, Bloomington, IN 47405, USA. Email:
[email protected]
2
1983, 1987; Hama and Chern, 1988; Hoynes and Schanzenbach, 2009; Johnson et al., 1981). Increased food expenditures related to FSP was observed in several food items including non-alcoholic beverages (Salathe, 1980). A recent study also revealed that SNAP was associated with reduced food insecurity, which may be related to improved health outcomes (Gundersen and Ziliak, 2015). Several studies have attempted to draw causal inferences of SNAP on food insecurity with observational data using a complicated modeling approach. Gundersen et al. (2017) used partial identification methods to estimate causal impact of SNAP on food insecurity, finding that SNAP was related to decreased prevalence of food insecurity. Analyzing data from the 2001–2006 National Health and Nutrition Examination Survey (NHANES), Kreider et al. (2012) revealed that SNAP participation was found to be associated with reduced food insecurity among children, when classification errors were addressed. Despite the considerable body of evidence suggesting the possibility that SNAP may predict nutrition well-being, there has been a growing movement to ban the purchases of sugar drinks with SNAP benefits. In 2010, New York City asked for the USDA’s permission to exclude sugar drinks containing more than 10 calories per 8-ounce serving from the list of SNAP-eligible items (Blondin, 2014). However, the USDA rejected the petition due to the concern that the New York City’s proposal would have been “too large and complex” to implement and evaluate (McGeehan, 2011). The mayors of New York, Los Angeles, Chicago, and 15 other cities sent a letter to congressional leaders in order to urge them to take legislative action on banning sugar drinks in SNAP, stating that “it is time to test and evaluate approaches limiting SNAP’s subsidization of products, such as sugar-sweetened beverages, that are contributing to obesity” (Becker et al., 2013). The consumption of sugar-sweetened soft drink (SSD) is positively associated with weight gain and type 2 diabetes (Apovian, 2004; Vartanian et al., 2007). In the US, there has been a considerable increase in the consumption of sugar drinks among adults since the late 1970s (Enns et al., 1997). In particular, low-income individuals appeared to consume more sugar drinks than high-income individuals. From 2005 to 2008, the mean percentage of total kilocalories consumed from sugar drinks was 8.8% among adults in households with income 130% of FPL might also receive SNAP benefits at the time of the survey due to their high levels of income volatility (Jolliffe and Ziliak, 2008; Todd and Ver Ploeg, 2014). Thus, we included low-income adults with a household income 250% of the FPL (N ¼ 1200) for this study. They were then categorized into two groups based on the household’s SNAP receipt: SNAP recipients (n ¼ 393) and non-recipients (n ¼ 807). In this study, respondents from the NHANES were not randomly assigned to SNAP recipients and non-recipients. Observable differences between treatment and control groups are more likely to occur in the non-randomized settings, which could lead to biased estimates of treatment effects (D’Agostino, 1998). Because traditional covariance adjustments may insufficient to eliminate this bias (D’Agostino, 1998), we used PSM to generate comparable SNAP recipients and non-recipients by balancing covariate distributions between the groups. PSM reduced the overall sample size from 1200 to 786 respondents, yielding the final sample of 393 matched pairs (SNAP recipients, n ¼ 393; non-recipients, n ¼ 393).
Measures The outcome variable for this study was frequency of SSD consumption. In the DSQ of the NHANES 2009–2010, respondents were asked “How often did you drink regular soda or pop that contains sugar in the past month? Do not include diet soda.” Responses for SSD consumption were measured as an ordinal variable by modifying frequency formats defined for the National Cancer Institute versions of the Diet History Questionnaire into three categories (National Cancer Institute, 2016): “1–3 times/month” coded as 1; “1–6 times/week” coded as 2; and “1 times/ day” coded as 3. SNAP participation was assessed based on the following three questions: “Have you or anyone in your household ever received FS (food stamp) benefits?” “In the last 12 months, did you or any member of your household receive FS benefits?” and “Number of days between the time the household last received FS benefits and the date of interview.” Respondents were classified as SNAP recipients if they received household FS benefits within 30 days at the time of survey. Respondents were classified as non-recipients if they did not receive household FS
3
benefits within the previous 12 months or if they had never received household FS benefits. Former SNAP recipients, defined as those who received household FS benefits within the past 12 months but did not receive them within 30 days at the time of survey, were excluded because of changes in their SNAP participation that might affect the consumption of SSD. Sociodemographic variables included age, gender (male, female), race/ethnicity (Hispanic, non-Hispanic White, non-Hispanic Black, others), highest level of education (high school education, college education), marital status (married or living with partner, formerly married, never married), household size, poverty-income ratio (calculated by dividing family income by the poverty guideline), and soda availability in homes (never/rarely, sometimes, most of times/always). These sociodemographic variables were included in the analysis as covariates because they were identified to be associated with SSD consumption (Hattersley et al., 2009; Rehm et al., 2008).
Statistical analysis Propensity-score matching (PSM) to generate the matched sample. The propensity score is a conditional probability that an individual will be a part of treatment group (i.e. SNAP recipients) based on his/her observed covariates (Rosenbaum and Rubin, 1983). The propensity score can be estimated using discriminant analysis or logistic regression if there is no missing value for covariates (D’Agostino, 1998). As a first step, we conducted logistic regression to estimate the propensity score by including respondents’ observed covariates (i.e. age, gender, race/ethnicity, education, marital status, household size, poverty-income ratio, and soda availability in homes). We then conducted an optimal pair matching to generate a matched sample. Matching defines a distance between treatment and control groups (i.e. SNAP recipients and non-recipients) and an algorithm is used to minimize the total distance between the groups (Ming and Rosenbaum, 2001). Optimal matching generally creates a matched sample where the total distance is minimized (Ming and Rosenbaum, 2001; Rosenbaum, 1989). Thus, we performed optimal pair matching using the “optmatch” package of R version 3.2.0. As a postmatching analysis, we conducted an imbalance test to determine if matching removed imbalance on each of the covariates between SNAP recipients and non-recipients. The absolute standardized mean difference in units of the pooled standard deviation was used as a criterion for assessing covariate balance between the groups (Austin, 2011). In general, it is expected that the absolute standardized mean difference for each of the covariates is reduced after matching. An imbalance test was conducted using Stata version 13.1. As a result, 393 matched pairs (SNAP recipients, n ¼ 393; non-recipients, n ¼ 393) were generated as the final sample.
4
Nutrition and Health XX(X)
Table 1. Results of imbalance test before and after propensityscore matching. Absolute Absolute standardized standardized difference before difference after matching matching Gender Age (year) Race Education Marital status Income-to-poverty ratio (%) Soda availability in home Household size
0.150522 0.181268 0.026373 0.243062 0.111436 0.738121 0.225412 0.199303
0.046425 0.008118 0.023564 0.065896 0.029459 0.072903 0.107457 0.008020
Note: before matching, N ¼ 1200 (SNAP recipients, n ¼ 393, nonrecipients, n ¼ 807); after matching, N ¼ 786 (SNAP recipients, n ¼ 393; non-recipients, n ¼ 393).
Ordinal logistic regression An ordinal logistic regression, using a proportional odds model, was conducted on the matched sample to determine the association of frequency of SSD consumption with SNAP participation, adjusting for all other covariates, such as gender, age, race/ethnicity, education, marital status, household size, poverty-income ratio, and soda availability in homes. The data were weighted using NHANES-defined weighting variables to account for the unequal probabilities of selection, non-response, and post-stratification adjustments. Odds ratios (ORs) and 95% CIs were generated using SAS version 9.4. This study was deemed exempt by the Institutional Review Board at Indiana University, Bloomington. This study was deemed exempt under federal regulation 45 46.101 (b) CFR.
Strengthening the Reporting of Observational studies in Epidemiology (STROBE) statement Our manuscript complied with the guidelines of the STROBE for cross-sectional studies to ensure the quality of observational study reporting (see Appendix 1).
Results Table 1 displays the results of the imbalance test, which illustrate the absolute standardized mean difference for each of the covariates before and after matching. The absolute standardized mean difference for each of the covariates was reduced after matching. It indicates that optimal pair matching balanced covariate distributions between SNAP recipients and non-recipients. Table 2 outlines the descriptive statistics of SNAP recipients and non-recipients before and after matching. Before matching, SNAP recipients significantly differed from non-recipients in terms of all covariates at the significance level of a ¼ 0.05. SNAP recipients had higher
proportions of female (55.2% vs. 47.7%), non-Hispanic White (41.7% vs. 36.4%), those with high school diploma or less (72% vs. 60.6%), and living in homes where soda is available most of times or always (68.2% vs. 58.9%), but lower proportions of married people (50.6% vs. 58.2%), compared with non-recipients. After matching, there were no longer significant differences on all observed covariates between SNAP recipients and non-recipients, suggesting that the matched SNAP recipients and non-recipients have similar characteristics. Table 3 displays the results of the ordinal logistic regression that determined the association of frequency of SSD consumption (“1–3 times/month,” “1–6 times/week,” and “1 times/day”) with SNAP participation, adjusting for age, gender, race/ethnicity, education, marital status, household size, poverty-income ratio, and soda availability in homes. As shown in the results of the adjusted model, SNAP recipients were more likely to report higher levels of SSD consumption, compared with non-recipients (adjusted odds ration (AOR) ¼ 1.55, 95% CI ¼ 1.172.07). Male gender, younger age, lower educational level, and soda availability in homes were also significantly associated with higher levels of SSD consumption among low-income adults. Male gender (AOR ¼ 1.69, 95% CI ¼ 1.172.46) and younger age (AOR ¼ 0.97, 95% CI ¼ 0.960.99) were positively associated with frequent SSD consumption. Individuals with high school education or less (AOR ¼ 2.28, 95% CI ¼ 1.333.89) were more likely to report higher levels of SSD consumption, compared with individuals with college education or more. Soda availability in homes was also positively associated with frequent SSD consumption (AOR ¼ 2.24, 95% CI ¼ 1.772.83).
Discussion This study used PSM to examine whether SNAP participation was associated with frequency of SSD consumption. We found that SNAP recipients consumed SSD more frequently, compared with non-recipients. Male gender, younger age, lower education level, and soda availability in homes were also identified to be associated with frequent SSD consumption among low-income adults. Our study findings underline the importance of formulating policies to reduce SSD consumption on the basis of this evidence. Our study revealed that frequency of SSD consumption differed by SNAP participation. Based on this finding, suggested recommendations may include restricting the accessibility of SSD with SNAP benefits to reduce SSD consumption. Limiting SSD in SNAP may also contribute to curbing obesity epidemics in the US. A recent study suggested that if the purchases of sugar drinks with SNAP benefits were banned, the prevalence of obesity and incidence of type 2 diabetes would be reduced, particularly among adults aged 18–65 years and some racial and ethnic minority groups (Basu et al., 2014). In addition, 510,000 diabetes person-years and 52,000 deaths from myocardial infarctions and strokes could be prevented over the next
Park et al.
5
Table 2. Descriptive statistics of study sample: adults (aged 20 years) in household 250% of the federal poverty level before and after propensity-score matching. Before matching (N ¼ 1200)
Covariates Gender: Male Female Age (year) [Mean (Std.)] Race: Hispanic Non-Hispanic White Non-Hispanic Black Other race including multi-racial Education: High school education College education Marital status: Married Formerly married Never married Income-to-poverty ratio (%) [Mean (Std.)] Household size [Mean (Std.)] Soda availability in home Never/rarely Sometimes Most of time/always Dependent variable SSD consumption: 1–3 times/month 1–6 times/week 1 times/day
Non-recipients (n ¼ 807)
SNAP recipients (n ¼ 393)
422 (52.3%) 385 (47.7%) 41.3 (14.3)
176 (44.8%) 217 (55.2%) 38.8 (13.2)
322 294 140 51
136 164 80 13
After matching (N ¼ 786) p-value
Non-recipients (n ¼ 393)
SNAP recipients (n ¼ 393)
185 (47.1%) 208 (52.9%) 38.7 (14.1)
176 (44.8%) 217 (55.2%) 38.8 (13.2)
163 135 65 30
136 164 80 13
0.02
(39.9%) (36.4%) (17.4%) (6.3%)
