Retailer Proximity and Nutrition Program Redemptions: Evidence From the Summer Electronic Benefit Transfer For Children Program

Retailer Proximity and Nutrition Program Redemptions: Evidence From the Summer Electronic Benefit... Abstract Although nearly all Supplemental Nutrition Assistance Program benefits are redeemed, a moderate share of Special Supplemental Nutrition Program for Women, Infant, and Children (WIC) benefits go unredeemed. Some hypothesize that the redemption rate differences are due to the lower density of WIC-authorized retailers. For the 2012 Summer Electronic Benefit Transfer for Children sites, this paper finds no consistent evidence of a relationship between redemption rates and retailer proximity. In fact, households often travel past the closest participating retailer to redeem their benefits. These findings are consistent with recent “food deserts” literature, which suggests that correlations between retailer environment and nutritional outcomes are not causal. Rather, it appears that differences in redemption rates may be related to the restrictions on what foods can be redeemed in what form with WIC. SNAP, WIC, benefit redemption, food access, geospatial analysis Redemption rates for the Supplemental Nutrition Assistance Program (SNAP) are much higher than for the Special Nutrition Assistance Program for Women Infants and Children (WIC) (Castner and Henke 2011). Fewer retailers participate in WIC than in SNAP. It is thus plausible that differences in households’ geographic proximity to participating retailers may explain some of this difference in redemption rates: if participating retailers are not readily accessible, households may redeem fewer of the benefits issued. For reasons discussed in the body of this paper, the Summer EBT for Children (SEBTC) Demonstration provides a valuable opportunity to explore distance to retailers as an explanation for incomplete redemption and the SNAP/WIC differential in redemption rates. Using data from participants in the 2012 SEBTC sites, analyses reported in this paper do not support the hypothesis that retailer proximity is the cause of variation in redemption rates between participants who received benefits through the SNAP versus WIC programs. The paper also reports results of some analyses consistent with the conjecture that more limited selection of WIC-allowable foods may explain that program’s lower redemption rates. The balance of this paper proceeds as follows. The following section provides context for the analyses with a brief review of the literature on food deserts. The next section describes the SEBTC Demonstration, the available data (including descriptive statistics), and why SEBTC provides a valuable opportunity to explore the relationship between proximity to retailers and redemption rates. The subsequent section reports results of regression analyses of how redemptions vary with distance to food retailers. The results do not support the hypothesis that longer average distances travelled to redeem WIC benefits contributes to lower rates of redemption in WIC compared to SNAP. Rather, the results suggest that redemption rates are only very weakly related to a household’s distance to closest participating supermarket/superstore or to any participating retailer. The following section reports results for where redemptions actually occur and their implications for explaining SNAP/WIC differences. The next section presents analyses of redemption rates by WIC food categories, consistent with the hypothesis that lower WIC redemption rates are related in part to a lack of interest in certain WIC foods among program participants. The final section summarizes the results and discusses their implications for both policy and broader food desert literature. Appendix A describes database construction in more detail and reports the results of some sensitivity analyses with respect to database construction choices. Appendix B details our methods and provides additional substantive results. Policy Background The USDA Food and Nutrition Service (USDA/FNS) provides food assistance directly to households through SNAP and WIC. These two national programs with similar goals differ in important ways. The larger of the two programs, SNAP, provides low-income households with resources to purchase almost any foods (the major exceptions are hot prepared foods, alcohol, and tobacco) at numerous food retailers. SNAP benefits are issued through an Electronic Benefit Transfer (EBT) card, which operates like a standard consumer debit card. WIC is more narrowly targeted to low-income women who are pregnant or who care for children up to age five. The WIC program is moving towards the distribution of benefits using an EBT card; some states are distributing WIC via EBT, while others are not. Unlike SNAP participants, WIC participants must have certification from a health care professional that they are at nutritional risk. WIC benefits are accepted at fewer retailers than are SNAP benefits and can be spent only on specified quantities of approved foods. The WIC food package for children comprises nine food categories: juice, milk, breakfast cereal, cheese, eggs, fruits and vegetables, whole wheat bread, canned fish, and legumes or peanut butter. Given the restrictions on what can be purchased and at which retailers, it is not surprising that national participation rates among eligible persons are lower for WIC than for SNAP. Indeed, 83% of persons meeting SNAP eligibility guidelines participate in the program versus 63% of those meeting WIC requirements. Participation in the WIC program is even lower for eligible children aged 1 to 4 years old (54%; Johnson et al. 2015; Gray and Cunnyngham 2016). In addition, the limited available evidence suggests that redemption rates among participants are also lower in WIC as compared to SNAP; WIC participants redeem only 83% of their total benefits while SNAP participants redeem 97% of their benefits (Phillips et al. 2014 for three states; USDA 2011 nationally). At least four major differences between the SNAP and WIC programs exist, which may account for the differences in redemption rates among WIC participants. First, the WIC eligible population is demographically narrower and a broader range of income are eligible. WIC is available only to households with an infant, child, or pregnant or post-partum woman, with incomes below 185% of the federal poverty line (and that meet other eligibility requirements; USDA FNS 2017a). SNAP is available to all households with income below 130% of the federal poverty line (and that meet other eligibility requirements; see USDA FNS 2017b). It seems plausible that households with lower incomes have a greater need and would therefore be more likely to redeem more benefits. It is possible that the WIC package is better-suited to some demographic groups, such that redemption rates would be higher or lower. Second, the WIC benefit is smaller. In 2015, the average monthly SNAP benefit per person was $126.81 (USDA FNS 2016a). Valuing the foods at cost, average monthly cost per enrolled person in WIC was estimated as $43.37 in 2015 (USDA FNS 2016b), approximately one-third of the average monthly SNAP benefit. It seems plausible that because the benefit is smaller, beneficiaries would be more likely to decide that it was not worth the effort to redeem the benefit (any, more, and all). Third, WIC can only be spent on its narrow set of foods. As noted, choice of foods is relatively unrestricted in SNAP; choice of foods is strictly limited in WIC, to only the approved foods in the food package. Some households may prefer not to consume some of these WIC foods. Household members may also experience confusion regarding which foods qualify. Qualifying foods may not be available. Each of these factors would lead to lower redemption rates. Fourth, WIC operates through a narrower retailer network. Fewer retailers accept WIC than accept SNAP. Perhaps transportation issues imply that the lower density of WIC retailers leads to lower redemption rates. Differences in retailer networks are the focus of this article, which also provides some evidence on the impact of the narrow choice of foods on redemption. The hypothesis that the narrower network of WIC retailers may explain differences between SNAP and WIC participation and redemption is related to the broader literature on “food deserts,” that is, areas with limited access to supermarkets. The food desert literature explores the relationship between food retailer access and a range of dietary outcomes (Larson, Story, and Nelson 2009). Simple correlations suggest that residing in a food desert is related to poorer nutritional outcomes (see the reviews in Larson, Story, and Nelson 2009, and the Institute of Medicine and National Research Council 2009; see also USDA 2009). Crucially, the idea of food deserts implicitly assumes that people purchase the majority of their food at local retailers. That implicit assumption appears to be incorrect. For low-income families who participate in food assistance programs, available evidence suggests that households travel moderate distances to shop, often passing closer retailers—and even passing closer supermarkets/superstores—to shop at more-distant retailers (Edin et al. 2013; Mabli and Worthington 2015; ver Ploeg et al. 2015; Grindal et al. 2016; Schwartz et al. 2017). The evidence suggests that low-income households redeem their nutrition benefits at retailers considerably farther away from their residence rather than the closest participating retailer or participating supermarket/superstore. This evidence makes it less plausible that residing in a food desert (i.e., the lack of a supermarket near a household’s residence) would affect nutritional outcomes. Consistent with this interpretation, both (a) more-formal causal analyses of the relation of proximity to retailers and nutritional intake and (b) more-detailed analyses of actual food shopping behavior suggest that the correlation of food deserts with poorer nutritional outcomes is not causal. That is, residing in a food desert does not seem to cause worse nutritional outcomes, and placing supermarkets in food deserts is unlikely to substantially improve nutrition (Kyureghian, Nayga, and Bhattacharya 2012; Cummins, Flint, and Matthews 2014; Alcott, Diamond, and Dube 2015; Dubowitz et al. 2015; Handbury, Rahkosky, and Schnell 2016; see also the review of the more recent literature in ver Ploeg and Wilde, forthcoming). Summer EBT for Children Demonstration Data for this study are drawn from the SEBTC evaluation. The SEBTC Demonstration was designed to improve children’s food security and nutritional status in the summer, when school-based meal programs operate on a much smaller scale. To address this lack of food assistance in selected demonstration sites, SEBTC provides nutritional assistance to families with children eligible for free and reduced-price meals during the school year. In conjunction with the demonstration, FNS funded a large random assignment evaluation. (See Collins et al. [2013] for details regarding the SEBTC Demonstration and results of the evaluation including a description of implementation and estimated impacts.) The demonstration and evaluation were implemented in the summers of 2011 through 2014 by 10 grantees in 16 communities. Because the number of participating sites and individuals vary over the course of the study and the study design changes slightly, this paper focuses on the 2012 sample only. The 2012 intervention involved the largest sample and the greatest number of sites (see Collins et al. [2013] for more on site selection and the sites themselves).1 Specifically, the program operated as follows. Potential sites applied to USDA/FNS. Selected sites were given the option of implementing SEBTC according to SNAP rules (SEBTC-SNAP) or according to WIC rules (SEBTC-WIC). In sites choosing SEBTC-SNAP, benefits were distributed through the SNAP EBT system and could be used to purchase almost any foods at SNAP retailers. Also, the benefit was set at $60 per participating child in the household. In sites choosing SEBTC-WIC, benefits were distributed through the WIC EBT system (only sites that had implemented WIC EBT could choose this option). The SEBTC-WIC package was specified to include foods appropriate for school-age children, with a value of approximately $60 per participating child in the household. Benefits could only be redeemed at WIC retailers (many SNAP retailers do not participate in WIC, but almost all WIC retailers participate in SNAP). Thus, SEBTC-SNAP versus SEBTC-WIC was not randomly assigned. However, because store locations and stores’ decisions to accept SNAP and/or WIC were determined independently of a site’s decision to administer SNAP or WIC benefits, we do not believe this influences our analysis. Households provided consent (active consent in some sites, passive consent in other sites) to participate in the study. Lists of children in consenting households were forwarded to the evaluation, which randomly assigned all children in a household to either treatment or control (i.e., in a single household either all eligible children were assigned to treatment or all eligible children were assigned to control). This paper analyzes households assigned to the treatment condition and a benefit valued at about $60 per eligible child. We do not use the data or outcomes of households who received no benefit. This contrast makes SEBTC a powerful context in which to understand differences in redemption rates of the SNAP and WIC programs. In the previous section, we noted four factors that vary between the SNAP and WIC programs: (a) different populations, (b) different benefit amounts, (c) more-restricted choice of foods, and (d) proximity of retailer. Simple SNAP-WIC comparisons mix the impact of all four considerations. In contrast, analyses of SEBTC essentially eliminate the first and second sources of variation (i.e., different populations and different benefit amounts). Across SEBTC-SNAP and SEBTC-WIC sites, the individual eligibility conditions are the same (children eligible for free and reduced-price school meals) as are the benefit amounts per eligible child.2 Thus, these comparisons of SEBTC-SNAP and SEBTC-WIC isolate the effect of the third and fourth sources of variation (i.e., more restricted choices of food and retailer proximity). This paper examines how redemption adjusts with variation in retailer proximity (the fourth reason for SNAP/WIC difference in redemption rates), measured by the distance to the closest supermarket or superstore where participants could redeem benefits. Any remaining differences are due to differences between the specific grantees or sites, restricted choice (the third explanation), or chance. Specifically, we construct our analysis file as follows. We begin with data on every household assigned to the treatment group and all SNAP and WIC retailers in each of the nine states that participated in SEBTC in 2013. We then geocode all households and all food retailers using ArcGIS Online Geocoding Service and compute the road network driving distance to the closest retailer—WIC, SNAP, any retailer, and supermarket/superstore, a total of four distances. As discussed in appendix A, we exclude households from the analyses for the following reasons: insufficient physical address data, excessive distance to closest retailer (more than 10 miles), and location within one mile of state border. Our data have few such households (733, or 2.1%). We also dropped two tribal sites (Cherokee and Chickasaw Nations) from the main analysis because a high proportion of households in these sites did not have physical addresses. In particular, the 10-mile cutoff was chosen to exclude only extreme observations. Few 2012 SEBTC households resided more than 10 miles from a participating retailer; from a total sample of about 30,000, the number of households that resided more than 10 miles from a participating retailer was only 115, and 73 of these households were in tribal sites. We analyzed the sensitivity of the paper’s results to these sample selection rules (see appendix B for details). Broadly speaking, the results are not sensitive to the inclusion of Cherokee and Chickasaw nation households or households that were excluded because of excessive distance to the nearest retailer or close proximity to the state border. In addition, we report site-specific results for the tribal sites. For analyses of redemption by retailer where SEBTC benefits were redeemed, we built a file with dollars redeemed by household by specific retailer. We then geocoded the distances from the household’s address to each of those retailers (see appendix A for details). Finally, for the restricted-choice analyses, we appended the dollar value of benefits redeemed in each WIC food category and analyzed the relationship to the nearest supermarket and redemption rates by food category. Table 2 presents sample statistics for the dependent variables. Considering the entire period in which the benefit was available in each site, redemption is nearly complete among SEBTC-SNAP households (94%) and substantially lower among SEBTC-WIC households (58%). Among those who redeemed any of their benefits (96% of SNAP households and 84% of WIC households), the average redemption rate was 88% overall, 98% for SNAP sites, and 69% for WIC sites. This finding of higher redemption rates for SEBTC-SNAP than for SEBTC-WIC is consistent with the finding of higher redemption rates for (conventional) SNAP than for (conventional) WIC. Table 1 Sample Statistics of Dependent Variables N Mean SD SNAP sites, 8 sites Redeemed any benefit 16,577 96.1% 19.5% Percent of benefit redeemed 16,577 93.9% 20.6% Percent of benefit redeemed (conditional on redeeming any benefit) 15,923 97.8% 8.0% Exhausted benefits in at least one month 16,577 81.1% 39.2% Exhausted benefits in all months 16,577 28.7% 45.2% WIC sites, 4 sites Redeemed any benefit 9,993 83.6% 37.0% Percent of benefit redeemed 9,993 57.8% 33.4% Percent of benefit redeemed (conditional on redeeming any benefit) 8,359 69.1% 23.5% Exhausted benefits in at least one month 9,993 13.3% 34.0% Exhausted benefits in all months 9,993 1.1% 10.3% N Mean SD SNAP sites, 8 sites Redeemed any benefit 16,577 96.1% 19.5% Percent of benefit redeemed 16,577 93.9% 20.6% Percent of benefit redeemed (conditional on redeeming any benefit) 15,923 97.8% 8.0% Exhausted benefits in at least one month 16,577 81.1% 39.2% Exhausted benefits in all months 16,577 28.7% 45.2% WIC sites, 4 sites Redeemed any benefit 9,993 83.6% 37.0% Percent of benefit redeemed 9,993 57.8% 33.4% Percent of benefit redeemed (conditional on redeeming any benefit) 8,359 69.1% 23.5% Exhausted benefits in at least one month 9,993 13.3% 34.0% Exhausted benefits in all months 9,993 1.1% 10.3% Source: SEBTC data. Table 1 Sample Statistics of Dependent Variables N Mean SD SNAP sites, 8 sites Redeemed any benefit 16,577 96.1% 19.5% Percent of benefit redeemed 16,577 93.9% 20.6% Percent of benefit redeemed (conditional on redeeming any benefit) 15,923 97.8% 8.0% Exhausted benefits in at least one month 16,577 81.1% 39.2% Exhausted benefits in all months 16,577 28.7% 45.2% WIC sites, 4 sites Redeemed any benefit 9,993 83.6% 37.0% Percent of benefit redeemed 9,993 57.8% 33.4% Percent of benefit redeemed (conditional on redeeming any benefit) 8,359 69.1% 23.5% Exhausted benefits in at least one month 9,993 13.3% 34.0% Exhausted benefits in all months 9,993 1.1% 10.3% N Mean SD SNAP sites, 8 sites Redeemed any benefit 16,577 96.1% 19.5% Percent of benefit redeemed 16,577 93.9% 20.6% Percent of benefit redeemed (conditional on redeeming any benefit) 15,923 97.8% 8.0% Exhausted benefits in at least one month 16,577 81.1% 39.2% Exhausted benefits in all months 16,577 28.7% 45.2% WIC sites, 4 sites Redeemed any benefit 9,993 83.6% 37.0% Percent of benefit redeemed 9,993 57.8% 33.4% Percent of benefit redeemed (conditional on redeeming any benefit) 8,359 69.1% 23.5% Exhausted benefits in at least one month 9,993 13.3% 34.0% Exhausted benefits in all months 9,993 1.1% 10.3% Source: SEBTC data. Table 2 Distance to Retailers Median Mean SD Min. 25th Percentile 75th Percentile Max. SNAP sites, 8 sites Driving distance to the nearest SNAP-accepting supermarket 1.00 1.28 1.14 0.00 0.66 1.47 9.96 Driving distance to the nearest SNAP-accepting retailer of any type 0.37 0.55 0.63 0.00 0.21 0.64 9.76 Driving distance to the nearest WIC-accepting supermarket 1.14 1.45 1.29 0.00 0.76 1.70 17.36 Driving distance to the nearest WIC-accepting retailer of any type 1.01 1.29 1.20 0.00 0.62 1.56 17.36 WIC sites, 4 sites Driving distance to the nearest SNAP-accepting supermarket 0.99 1.48 1.62 0.01 0.61 1.59 10.00 Driving distance to the nearest SNAP-accepting retailer of any type 0.38 0.63 0.87 0.00 0.21 0.67 9.06 Driving distance to the nearest WIC-accepting supermarket 1.05 1.56 1.65 0.01 0.64 1.70 10.00 Driving distance to the nearest WIC-accepting retailer of any type 0.91 1.36 1.53 0.00 0.46 1.59 9.98 Median Mean SD Min. 25th Percentile 75th Percentile Max. SNAP sites, 8 sites Driving distance to the nearest SNAP-accepting supermarket 1.00 1.28 1.14 0.00 0.66 1.47 9.96 Driving distance to the nearest SNAP-accepting retailer of any type 0.37 0.55 0.63 0.00 0.21 0.64 9.76 Driving distance to the nearest WIC-accepting supermarket 1.14 1.45 1.29 0.00 0.76 1.70 17.36 Driving distance to the nearest WIC-accepting retailer of any type 1.01 1.29 1.20 0.00 0.62 1.56 17.36 WIC sites, 4 sites Driving distance to the nearest SNAP-accepting supermarket 0.99 1.48 1.62 0.01 0.61 1.59 10.00 Driving distance to the nearest SNAP-accepting retailer of any type 0.38 0.63 0.87 0.00 0.21 0.67 9.06 Driving distance to the nearest WIC-accepting supermarket 1.05 1.56 1.65 0.01 0.64 1.70 10.00 Driving distance to the nearest WIC-accepting retailer of any type 0.91 1.36 1.53 0.00 0.46 1.59 9.98 Source: SEBTC data, FNS. Table 2 Distance to Retailers Median Mean SD Min. 25th Percentile 75th Percentile Max. SNAP sites, 8 sites Driving distance to the nearest SNAP-accepting supermarket 1.00 1.28 1.14 0.00 0.66 1.47 9.96 Driving distance to the nearest SNAP-accepting retailer of any type 0.37 0.55 0.63 0.00 0.21 0.64 9.76 Driving distance to the nearest WIC-accepting supermarket 1.14 1.45 1.29 0.00 0.76 1.70 17.36 Driving distance to the nearest WIC-accepting retailer of any type 1.01 1.29 1.20 0.00 0.62 1.56 17.36 WIC sites, 4 sites Driving distance to the nearest SNAP-accepting supermarket 0.99 1.48 1.62 0.01 0.61 1.59 10.00 Driving distance to the nearest SNAP-accepting retailer of any type 0.38 0.63 0.87 0.00 0.21 0.67 9.06 Driving distance to the nearest WIC-accepting supermarket 1.05 1.56 1.65 0.01 0.64 1.70 10.00 Driving distance to the nearest WIC-accepting retailer of any type 0.91 1.36 1.53 0.00 0.46 1.59 9.98 Median Mean SD Min. 25th Percentile 75th Percentile Max. SNAP sites, 8 sites Driving distance to the nearest SNAP-accepting supermarket 1.00 1.28 1.14 0.00 0.66 1.47 9.96 Driving distance to the nearest SNAP-accepting retailer of any type 0.37 0.55 0.63 0.00 0.21 0.64 9.76 Driving distance to the nearest WIC-accepting supermarket 1.14 1.45 1.29 0.00 0.76 1.70 17.36 Driving distance to the nearest WIC-accepting retailer of any type 1.01 1.29 1.20 0.00 0.62 1.56 17.36 WIC sites, 4 sites Driving distance to the nearest SNAP-accepting supermarket 0.99 1.48 1.62 0.01 0.61 1.59 10.00 Driving distance to the nearest SNAP-accepting retailer of any type 0.38 0.63 0.87 0.00 0.21 0.67 9.06 Driving distance to the nearest WIC-accepting supermarket 1.05 1.56 1.65 0.01 0.64 1.70 10.00 Driving distance to the nearest WIC-accepting retailer of any type 0.91 1.36 1.53 0.00 0.46 1.59 9.98 Source: SEBTC data, FNS. Table 2 presents descriptive statistics for the key independent variable, distance to closest retailer in various classes. Distance (in miles) to closest participating retailer—SNAP retailers in SEBTC-SNAP sites and WIC retailers in SEBTC-WIC sites—is shorter for SNAP sites (0.37 median, 0.55 mean) than for WIC sites (0.91 median, 1.36 mean). Differences are similar but less stark for distance to a participating supermarket or superstore (1.00 median, 1.28 mean for SEBTC-SNAP versus 1.05 median, 1.56 mean for SEBTC-WIC). Because distance is not randomly assigned and we are interested in plausibly causal interpretation of the analyses, we control for the following household characteristics: number of children in the household, age of the oldest child, race/ethnicity of the head of household, free or reduced lunch price eligibility, and household familial structure. We also include census tract-level regressors, including population density, median household income, population with a high school degree, population with no access to a vehicle, and population below the federal poverty line. We consider four specifications: (a) no controls (i.e., regressors), (b) site indicators, (c) site indicators, plus household characteristics, and (d) site indicators, characteristics of the sample household itself, and households’ characteristics for the census tract (computed from the five-year American Community Survey tabulations for 2012). Table 3 lists these controls and their sample statistics. Table 3 Sample Statistics of Household Covariates Household N Mean SD SNAP sites, 8 sites Household characteristics Number of children in household 16,577 1.8 1.0 Head of household is Black, non-Hispanic 11,654 46.0% 49.8% Head of household is Hispanic 11,654 11.7% 32.2% Household children are eligible for free lunch 13,235 91.9% 27.3% Household children are eligible for reduced-price lunch 13,235 9.7% 29.5% Oldest child is under 21 12,136 100.0% 1.6% Household is headed by a single female 13,728 33.7% 47.3% Household is headed by a single male 13,728 3.7% 19.0% Census tract characteristics Population density (persons per square mile) 16,571 4,524 3,956 Median income 16,571 $43,824 $18,493 Population with high school degree 16,571 82.0% 9.7% Population with no access to a vehicle 16,571 14.2% 12.2% Population below the federal poverty line 16,571 22.3% 13.8% WIC sites, 4 sites Household characteristics Number of children in household 9,993 1.8 1.0 Head of household is Black, non-Hispanic 7,893 10.4% 30.6% Head of household is Hispanic 7,893 48.7% 50.0% Household children are eligible for free lunch 5,027 91.0% 28.6% Household children are eligible for reduced-price lunch 5,027 11.5% 31.9% Oldest child is under 21 7,911 99.5% 7.1% Household is headed by a single female 7,913 20.3% 40.2% Household is headed by a single male 7,913 4.7% 21.2% Census tract characteristics Population density (persons per square mile) 9,993 4,555 3,132 Median income 9,989 $39,192 $14,181 Population with high school degree 9,993 76.5% 13.9% Population with no access to a vehicle 9,993 10.2% 7.8% Population below the federal poverty line 9,993 25.4% 13.7% Household N Mean SD SNAP sites, 8 sites Household characteristics Number of children in household 16,577 1.8 1.0 Head of household is Black, non-Hispanic 11,654 46.0% 49.8% Head of household is Hispanic 11,654 11.7% 32.2% Household children are eligible for free lunch 13,235 91.9% 27.3% Household children are eligible for reduced-price lunch 13,235 9.7% 29.5% Oldest child is under 21 12,136 100.0% 1.6% Household is headed by a single female 13,728 33.7% 47.3% Household is headed by a single male 13,728 3.7% 19.0% Census tract characteristics Population density (persons per square mile) 16,571 4,524 3,956 Median income 16,571 $43,824 $18,493 Population with high school degree 16,571 82.0% 9.7% Population with no access to a vehicle 16,571 14.2% 12.2% Population below the federal poverty line 16,571 22.3% 13.8% WIC sites, 4 sites Household characteristics Number of children in household 9,993 1.8 1.0 Head of household is Black, non-Hispanic 7,893 10.4% 30.6% Head of household is Hispanic 7,893 48.7% 50.0% Household children are eligible for free lunch 5,027 91.0% 28.6% Household children are eligible for reduced-price lunch 5,027 11.5% 31.9% Oldest child is under 21 7,911 99.5% 7.1% Household is headed by a single female 7,913 20.3% 40.2% Household is headed by a single male 7,913 4.7% 21.2% Census tract characteristics Population density (persons per square mile) 9,993 4,555 3,132 Median income 9,989 $39,192 $14,181 Population with high school degree 9,993 76.5% 13.9% Population with no access to a vehicle 9,993 10.2% 7.8% Population below the federal poverty line 9,993 25.4% 13.7% Source: SEBTC data and appended American Community Survey census tract characteristics data. Table 3 Sample Statistics of Household Covariates Household N Mean SD SNAP sites, 8 sites Household characteristics Number of children in household 16,577 1.8 1.0 Head of household is Black, non-Hispanic 11,654 46.0% 49.8% Head of household is Hispanic 11,654 11.7% 32.2% Household children are eligible for free lunch 13,235 91.9% 27.3% Household children are eligible for reduced-price lunch 13,235 9.7% 29.5% Oldest child is under 21 12,136 100.0% 1.6% Household is headed by a single female 13,728 33.7% 47.3% Household is headed by a single male 13,728 3.7% 19.0% Census tract characteristics Population density (persons per square mile) 16,571 4,524 3,956 Median income 16,571 $43,824 $18,493 Population with high school degree 16,571 82.0% 9.7% Population with no access to a vehicle 16,571 14.2% 12.2% Population below the federal poverty line 16,571 22.3% 13.8% WIC sites, 4 sites Household characteristics Number of children in household 9,993 1.8 1.0 Head of household is Black, non-Hispanic 7,893 10.4% 30.6% Head of household is Hispanic 7,893 48.7% 50.0% Household children are eligible for free lunch 5,027 91.0% 28.6% Household children are eligible for reduced-price lunch 5,027 11.5% 31.9% Oldest child is under 21 7,911 99.5% 7.1% Household is headed by a single female 7,913 20.3% 40.2% Household is headed by a single male 7,913 4.7% 21.2% Census tract characteristics Population density (persons per square mile) 9,993 4,555 3,132 Median income 9,989 $39,192 $14,181 Population with high school degree 9,993 76.5% 13.9% Population with no access to a vehicle 9,993 10.2% 7.8% Population below the federal poverty line 9,993 25.4% 13.7% Household N Mean SD SNAP sites, 8 sites Household characteristics Number of children in household 16,577 1.8 1.0 Head of household is Black, non-Hispanic 11,654 46.0% 49.8% Head of household is Hispanic 11,654 11.7% 32.2% Household children are eligible for free lunch 13,235 91.9% 27.3% Household children are eligible for reduced-price lunch 13,235 9.7% 29.5% Oldest child is under 21 12,136 100.0% 1.6% Household is headed by a single female 13,728 33.7% 47.3% Household is headed by a single male 13,728 3.7% 19.0% Census tract characteristics Population density (persons per square mile) 16,571 4,524 3,956 Median income 16,571 $43,824 $18,493 Population with high school degree 16,571 82.0% 9.7% Population with no access to a vehicle 16,571 14.2% 12.2% Population below the federal poverty line 16,571 22.3% 13.8% WIC sites, 4 sites Household characteristics Number of children in household 9,993 1.8 1.0 Head of household is Black, non-Hispanic 7,893 10.4% 30.6% Head of household is Hispanic 7,893 48.7% 50.0% Household children are eligible for free lunch 5,027 91.0% 28.6% Household children are eligible for reduced-price lunch 5,027 11.5% 31.9% Oldest child is under 21 7,911 99.5% 7.1% Household is headed by a single female 7,913 20.3% 40.2% Household is headed by a single male 7,913 4.7% 21.2% Census tract characteristics Population density (persons per square mile) 9,993 4,555 3,132 Median income 9,989 $39,192 $14,181 Population with high school degree 9,993 76.5% 13.9% Population with no access to a vehicle 9,993 10.2% 7.8% Population below the federal poverty line 9,993 25.4% 13.7% Source: SEBTC data and appended American Community Survey census tract characteristics data. Table 4 Effects of Distance to the Closest Supermarket on Proportion Redeemed All Households SNAP Households WIC Households Driving distance to the nearest supermarket where benefits could be redeemed 0.000 −0.001 0.001 (0.001) (0.001) (0.002) Constant 0.605*** 0.878*** 0.680*** (0.050) (0.066) (0.074) R-squared 0.345 0.026 0.075 N 26,570 16,577 9,993 All Households SNAP Households WIC Households Driving distance to the nearest supermarket where benefits could be redeemed 0.000 −0.001 0.001 (0.001) (0.001) (0.002) Constant 0.605*** 0.878*** 0.680*** (0.050) (0.066) (0.074) R-squared 0.345 0.026 0.075 N 26,570 16,577 9,993 Note: Two-sided hypothesis test; asterisks *** = statistically significant p<0.01; ** = statistically significant, p<0.05; and * = statistically significant, p<0.10. Table 4 Effects of Distance to the Closest Supermarket on Proportion Redeemed All Households SNAP Households WIC Households Driving distance to the nearest supermarket where benefits could be redeemed 0.000 −0.001 0.001 (0.001) (0.001) (0.002) Constant 0.605*** 0.878*** 0.680*** (0.050) (0.066) (0.074) R-squared 0.345 0.026 0.075 N 26,570 16,577 9,993 All Households SNAP Households WIC Households Driving distance to the nearest supermarket where benefits could be redeemed 0.000 −0.001 0.001 (0.001) (0.001) (0.002) Constant 0.605*** 0.878*** 0.680*** (0.050) (0.066) (0.074) R-squared 0.345 0.026 0.075 N 26,570 16,577 9,993 Note: Two-sided hypothesis test; asterisks *** = statistically significant p<0.01; ** = statistically significant, p<0.05; and * = statistically significant, p<0.10. Analyses discussed in the body of the paper use the full set of controls. Appendix B explores the sensitivity of the paper’s main results to the specific controls included in the regression models. Once the site indicators are included (i.e., the second specification), we observe only minimal sensitivity (see appendix table B.1.1). The body of the article therefore only reports results for the fourth specification. Redemption and Distance to Retailers This section uses regression analysis to explore the relationship between distance to closest retailer and redemption rate, that is, benefits redeemed as a percentage of benefits issued. Analyses in appendix B show results for similar models for redemption of both any and all benefits. General patterns are similar. The generic regression is of the form: Y = β0+ β1Distance + β2ControlVars + β3Site +u (1) where Y represents our benefit redemption outcomes; Distance represents distance to some category of food retailer where a household could have redeemed its SEBTC benefits;3 in the primary specification, this is the distance to the nearest SNAP-accepting supermarket (SNAP sites) and to the nearest WIC-accepting supermarket (WIC sites); ControlVars is a vector of characteristics from the SEBTC sampling file and census tract; Site represents a dummy variable for each demonstration site (=1 for the site in which the household was located, and 0 otherwise); β0, β1, β2, and β3, are parameters to be estimated; u is a regression residual. For those analyses in which the benefit redemption outcomes are binary (e.g., take-up of any benefit and benefit exhaustion) these analyses are linear probability models. We run all analyses separately for WIC and SNAP sites. Table 4 shows the results of the regression analyses for SNAP households and WIC households separately, where “distance” is miles. These results imply that there is virtually no relationship between distance to the nearest supermarket where benefits could be redeemed and the proportion of benefits redeemed over the course of the summer of 2012. For SNAP sites there is a small negative (but not statistically significant) relationship between distance and benefits redeemed. A SNAP household that is 10 miles away from its nearest SNAP-accepting supermarket (the largest distance in our analysis file) would, on average, redeem 1 percentage point less of its benefits. The opposite is true when including only WIC households: a WIC household 10 miles4 away from the closest supermarket that accepted WIC benefits would, on average, redeem 1 percentage point more of its total benefits. Furthermore, the results are precise: 95% confidence intervals for SNAP sites are approximately -3% to -1%, and for WIC sites approximately −1% to 3%. Results are not sensitive to covariates (table B.1.1) for distance to closest retailer (rather than closest supermarket or superstore table B.1.2) and when including both distance to closest supermarket/superstore and to any retailer (table B.1.3). The previous model assumes that the change in redemption is constant with each additional mile to the closest retailer. To relax this assumption, we estimated linear spline models, which allow the effect of the first mile (or fraction thereof) to differ from the effect of miles past the first mile (Grindal et al. [2016] find some evidence for such a difference.) We also consider models in which the break point is at three-quarters of a mile (rather than a mile). Estimates from these spline models imply that redemption decreases slightly (and the relationship is statistically significant) for the first mile between the household and the retailer (about 1 percentage point), but there is no relationship if instead the knot is at 0.75 miles (see appendix table B.1.4). Considering each of the 12 participating sites separately, in a linear model, the effects of distance to the nearest supermarket are near 0% in each site (table 5). In some sites the effect is slightly negative, and in others it is slightly positive. In Delaware (a SNAP site) and Grand Rapids (a WIC site) we observe significant, negative relationships between distance and redemption rates. In Delaware, a participant redeemed 0.7% less of his or her benefits for each additional mile between a household and the nearest supermarket where they could redeem benefits. In Grand Rapids, participants, on average, redeemed nearly 2% less of their benefit amount for each additional mile. Again, these site-specific results should be interpreted with caution given the numerous parameters estimated. Perhaps distance matters in some sites, but not others. Alternatively, the results may be due to chance, as we see no conclusive patterns or site characteristics that would allow us to explain variations in our findings. Distance to Retailers Where Benefits Are Redeemed In the summer of 2012, there were approximately 50,000 authorized SNAP retailers in the nine states where the SEBTC sites were located. Only 7,186 of these retailers also accepted WIC (15%). Despite the large number of retailers who accept SNAP benefits, nearly two-thirds of SEBTC-SNAP redemptions (65% of dollars redeemed) in July of 2012 (the month when all demonstration sites participated) were redeemed at stores that also accepted WIC. This high proportion of redemptions in SEBTC-SNAP sites at WIC-accepting retailers is probably related in part to the fact that the overwhelming share of benefits are spent at supermarkets and superstores (86% of SEBTC redemptions in SNAP sites and 89% of SEBTC redemptions in WIC sites), and most supermarkets and superstores accept both WIC and SNAP. Of all SNAP-accepting supermarkets and superstores, 69% of SEBTC redemptions occur at supermarkets and superstores that accept WIC. Few SEBTC benefits were redeemed at participants’ closest participating retailer or even the closest participating supermarket. In WIC sites, 29% of redemptions were made at a household’s closest participating retailer (i.e., where any redemption was made), and 25% of redemptions were made at a household’s closest participating supermarket (sometimes the closest participating retailer is a supermarket). In SNAP sites, only 4% of redemptions were made at the closest participating retailer, and only 2% at the closest participating supermarket. Furthermore, it is not just that beneficiaries do not redeem at the closest retailer. In SNAP sites, the majority of redemptions (53%) were made at stores farther away than the 25 closest participating retailers. In both SNAP and WIC sites, participants are traveling approximately four times farther than the closest supermarket or superstore where they could redeem benefits. Weighting by benefits redeemed, the average distance traveled was 3.3 miles for an SEBTC-SNAP household and 4.6 miles for an SEBTC-WIC household. Similarly, the distance to the retailer at which the most benefits were redeemed was 3.9 miles for SEBTC-SNAP and 4.7 miles for SEBTC-WIC. Each of these distances is several times the mean distance to the closest participating retailer (0.6 miles for SEBTC-SNAP and 1.4 miles for SEBTC-WIC) or the closest participating supermarket/superstore (1.3 miles for SEBTC-SNAP and 1.6 miles for SEBTC-WIC). Results on Redemption by Specific Foods WIC participants can redeem benefits only for approved foods within the specific categories and at quantities of their food prescriptions. One hypothesis for lower redemption rates in WIC sites is that participants may not want to redeem some (or all of some) foods, even if they are available with minimal travel. It is either not perceived as worth the effort of carrying them home or the inconvenience of separating their purchases, or there is so little interest in the allowed foods. Other hypotheses are possible. Some of the foods are more perishable (milk, cheese, eggs, fruits and vegetables); others are less perishable (cereal, beans, canned fish). Because of the details of allowable WIC foods within a category, some foods are more challenging to obtain in acceptable forms (cereal, bread, juice) and in quantities that allow spending out a category. Table 6 shows the average dollar amounts spent in each category for each of the four WIC sites included in the analysis. Prices of groceries vary from site to site, and allowable quantities vary across food categories, so it is more informative to look at the within-category redemption rates. Table 7 shows the redemption rates for each WIC food category, calculated as the total cost of units purchased/total cost of units allowed for all foods covered by the WIC benefit in a specific category. Considering all WIC sites together and each site separately, redemption rates were generally lowest in the beans, fish, and bread categories. However, excluding households that redeemed none of their SEBTC-WIC benefits, redemption rates still do not approach the rates of nearly complete redemptions in the SNAP sites. Table 5 Effects of Distance to the Closest Supermarket/Superstore where Benefits Could Be Redeemed on Percentage Redeemed by Site Coefficient (SE) SNAP Sites Eastern Connecticut 0.003 (0.002) Western Connecticut 0.016 (0.012) Delaware −0.007* (0.004) Missouri-Kansas City −0.007 (0.010) Missouri-St. Louis −0.012 (0.010) Oregon-Deschutes −0.001 (0.002) Oregon-Salem 0.002 (0.004) Washington −0.001 (0.006) WIC Sites Cherokee −0.000 (0.004) Chickasaw 0.004 (0.003) Michigan-Eastern 0.004 (0.002) Michigan-Grand Rapids −0.019* (0.010) Nevada −0.000 (0.004) Texas −0.006 (0.010) Coefficient (SE) SNAP Sites Eastern Connecticut 0.003 (0.002) Western Connecticut 0.016 (0.012) Delaware −0.007* (0.004) Missouri-Kansas City −0.007 (0.010) Missouri-St. Louis −0.012 (0.010) Oregon-Deschutes −0.001 (0.002) Oregon-Salem 0.002 (0.004) Washington −0.001 (0.006) WIC Sites Cherokee −0.000 (0.004) Chickasaw 0.004 (0.003) Michigan-Eastern 0.004 (0.002) Michigan-Grand Rapids −0.019* (0.010) Nevada −0.000 (0.004) Texas −0.006 (0.010) Note: Two-sided hypothesis test; asterisks *** = statistically significant, p<0.01; ** = statistically significant, p<0.05; * = statistically significant, p<0.10. Table 5 Effects of Distance to the Closest Supermarket/Superstore where Benefits Could Be Redeemed on Percentage Redeemed by Site Coefficient (SE) SNAP Sites Eastern Connecticut 0.003 (0.002) Western Connecticut 0.016 (0.012) Delaware −0.007* (0.004) Missouri-Kansas City −0.007 (0.010) Missouri-St. Louis −0.012 (0.010) Oregon-Deschutes −0.001 (0.002) Oregon-Salem 0.002 (0.004) Washington −0.001 (0.006) WIC Sites Cherokee −0.000 (0.004) Chickasaw 0.004 (0.003) Michigan-Eastern 0.004 (0.002) Michigan-Grand Rapids −0.019* (0.010) Nevada −0.000 (0.004) Texas −0.006 (0.010) Coefficient (SE) SNAP Sites Eastern Connecticut 0.003 (0.002) Western Connecticut 0.016 (0.012) Delaware −0.007* (0.004) Missouri-Kansas City −0.007 (0.010) Missouri-St. Louis −0.012 (0.010) Oregon-Deschutes −0.001 (0.002) Oregon-Salem 0.002 (0.004) Washington −0.001 (0.006) WIC Sites Cherokee −0.000 (0.004) Chickasaw 0.004 (0.003) Michigan-Eastern 0.004 (0.002) Michigan-Grand Rapids −0.019* (0.010) Nevada −0.000 (0.004) Texas −0.006 (0.010) Note: Two-sided hypothesis test; asterisks *** = statistically significant, p<0.01; ** = statistically significant, p<0.05; * = statistically significant, p<0.10. Table 6 Average Dollar Amounts Redeemed, in Each WIC Category, over the Entire Summer Site N Total Bread Cheese Beans Cereal Eggs Fish Fruits/veg Juice Milk Michigan - Eastern 1,390 $438.36 $32.06 $37.29 $42.54 $49.75 $11.16 $16.11 $106.13 $23.48 $51.53 Michigan – Grand Rapids 2,323 $327.64 $20.15 $24.53 $19.81 $36.27 $7.25 $10.84 $71.71 $16.33 $41.24 NV 2,383 $375.48 $30.35 $18.40 $15.41 $28.77 $6.67 $11.87 $61.87 $12.81 $36.82 TX 2,421 $274.94 $22.62 $18.66 $19.79 $28.65 $7.53 $13.21 $67.49 10.86 40.45 Site N Total Bread Cheese Beans Cereal Eggs Fish Fruits/veg Juice Milk Michigan - Eastern 1,390 $438.36 $32.06 $37.29 $42.54 $49.75 $11.16 $16.11 $106.13 $23.48 $51.53 Michigan – Grand Rapids 2,323 $327.64 $20.15 $24.53 $19.81 $36.27 $7.25 $10.84 $71.71 $16.33 $41.24 NV 2,383 $375.48 $30.35 $18.40 $15.41 $28.77 $6.67 $11.87 $61.87 $12.81 $36.82 TX 2,421 $274.94 $22.62 $18.66 $19.79 $28.65 $7.53 $13.21 $67.49 10.86 40.45 Table 6 Average Dollar Amounts Redeemed, in Each WIC Category, over the Entire Summer Site N Total Bread Cheese Beans Cereal Eggs Fish Fruits/veg Juice Milk Michigan - Eastern 1,390 $438.36 $32.06 $37.29 $42.54 $49.75 $11.16 $16.11 $106.13 $23.48 $51.53 Michigan – Grand Rapids 2,323 $327.64 $20.15 $24.53 $19.81 $36.27 $7.25 $10.84 $71.71 $16.33 $41.24 NV 2,383 $375.48 $30.35 $18.40 $15.41 $28.77 $6.67 $11.87 $61.87 $12.81 $36.82 TX 2,421 $274.94 $22.62 $18.66 $19.79 $28.65 $7.53 $13.21 $67.49 10.86 40.45 Site N Total Bread Cheese Beans Cereal Eggs Fish Fruits/veg Juice Milk Michigan - Eastern 1,390 $438.36 $32.06 $37.29 $42.54 $49.75 $11.16 $16.11 $106.13 $23.48 $51.53 Michigan – Grand Rapids 2,323 $327.64 $20.15 $24.53 $19.81 $36.27 $7.25 $10.84 $71.71 $16.33 $41.24 NV 2,383 $375.48 $30.35 $18.40 $15.41 $28.77 $6.67 $11.87 $61.87 $12.81 $36.82 TX 2,421 $274.94 $22.62 $18.66 $19.79 $28.65 $7.53 $13.21 $67.49 10.86 40.45 Table 7 Average Redemption Rates by Food Category All WIC Sites Eastern Michigan Grand Rapids, Michigan Nevada Texas Unconditional Conditional Unconditional Conditional Unconditional Conditional Unconditional Conditional Unconditional Conditional N 9,988 8,359 1,422 1,350 2,773 2,532 2,887 2,346 2,906 2,131 Milk 65.8% 78.7% 75.8% 79.9% 70.3% 76.9% 57.8% 71.1% 64.7% 88.2% Cheese 66.3% 79.3% 76.6% 80.7% 68.7% 75.3% 58.0% 71.3% 67.4% 91.9% Eggs 66.1% 79.0% 74.2% 78.2% 75.1% 82.3% 51.5% 63.3% 68.1% 92.9% Juice 66.0% 78.9% 74.4% 78.4% 73.7% 80.7% 53.8% 66.2% 66.7% 90.9% Cereal 60.6% 72.4% 64.3% 67.8% 62.4% 68.3% 52.0% 64.0% 65.6% 89.5% Beans 53.9% 64.3% 67.9% 71.5% 55.7% 61.0% 37.4% 46.0% 61.6% 84.0% Fish 52.1% 62.2% 70.6% 74.3% 57.2% 62.7% 33.2% 40.9% 56.9% 77.6% Bread 48.1% 57.5% 42.0% 44.2% 41.9% 45.9% 43.0% 52.9% 62.2% 84.8% Fruits/ veg 64.4% 76.9% 68.1% 71.8% 72.3% 79.2% 54.7% 67.3% 64.5% 88.0% All WIC Sites Eastern Michigan Grand Rapids, Michigan Nevada Texas Unconditional Conditional Unconditional Conditional Unconditional Conditional Unconditional Conditional Unconditional Conditional N 9,988 8,359 1,422 1,350 2,773 2,532 2,887 2,346 2,906 2,131 Milk 65.8% 78.7% 75.8% 79.9% 70.3% 76.9% 57.8% 71.1% 64.7% 88.2% Cheese 66.3% 79.3% 76.6% 80.7% 68.7% 75.3% 58.0% 71.3% 67.4% 91.9% Eggs 66.1% 79.0% 74.2% 78.2% 75.1% 82.3% 51.5% 63.3% 68.1% 92.9% Juice 66.0% 78.9% 74.4% 78.4% 73.7% 80.7% 53.8% 66.2% 66.7% 90.9% Cereal 60.6% 72.4% 64.3% 67.8% 62.4% 68.3% 52.0% 64.0% 65.6% 89.5% Beans 53.9% 64.3% 67.9% 71.5% 55.7% 61.0% 37.4% 46.0% 61.6% 84.0% Fish 52.1% 62.2% 70.6% 74.3% 57.2% 62.7% 33.2% 40.9% 56.9% 77.6% Bread 48.1% 57.5% 42.0% 44.2% 41.9% 45.9% 43.0% 52.9% 62.2% 84.8% Fruits/ veg 64.4% 76.9% 68.1% 71.8% 72.3% 79.2% 54.7% 67.3% 64.5% 88.0% Note: “Unconditional”—among everyone, whether or not any benefits were redeemed; “Conditional”—among those who redeemed any benefits. Table 7 Average Redemption Rates by Food Category All WIC Sites Eastern Michigan Grand Rapids, Michigan Nevada Texas Unconditional Conditional Unconditional Conditional Unconditional Conditional Unconditional Conditional Unconditional Conditional N 9,988 8,359 1,422 1,350 2,773 2,532 2,887 2,346 2,906 2,131 Milk 65.8% 78.7% 75.8% 79.9% 70.3% 76.9% 57.8% 71.1% 64.7% 88.2% Cheese 66.3% 79.3% 76.6% 80.7% 68.7% 75.3% 58.0% 71.3% 67.4% 91.9% Eggs 66.1% 79.0% 74.2% 78.2% 75.1% 82.3% 51.5% 63.3% 68.1% 92.9% Juice 66.0% 78.9% 74.4% 78.4% 73.7% 80.7% 53.8% 66.2% 66.7% 90.9% Cereal 60.6% 72.4% 64.3% 67.8% 62.4% 68.3% 52.0% 64.0% 65.6% 89.5% Beans 53.9% 64.3% 67.9% 71.5% 55.7% 61.0% 37.4% 46.0% 61.6% 84.0% Fish 52.1% 62.2% 70.6% 74.3% 57.2% 62.7% 33.2% 40.9% 56.9% 77.6% Bread 48.1% 57.5% 42.0% 44.2% 41.9% 45.9% 43.0% 52.9% 62.2% 84.8% Fruits/ veg 64.4% 76.9% 68.1% 71.8% 72.3% 79.2% 54.7% 67.3% 64.5% 88.0% All WIC Sites Eastern Michigan Grand Rapids, Michigan Nevada Texas Unconditional Conditional Unconditional Conditional Unconditional Conditional Unconditional Conditional Unconditional Conditional N 9,988 8,359 1,422 1,350 2,773 2,532 2,887 2,346 2,906 2,131 Milk 65.8% 78.7% 75.8% 79.9% 70.3% 76.9% 57.8% 71.1% 64.7% 88.2% Cheese 66.3% 79.3% 76.6% 80.7% 68.7% 75.3% 58.0% 71.3% 67.4% 91.9% Eggs 66.1% 79.0% 74.2% 78.2% 75.1% 82.3% 51.5% 63.3% 68.1% 92.9% Juice 66.0% 78.9% 74.4% 78.4% 73.7% 80.7% 53.8% 66.2% 66.7% 90.9% Cereal 60.6% 72.4% 64.3% 67.8% 62.4% 68.3% 52.0% 64.0% 65.6% 89.5% Beans 53.9% 64.3% 67.9% 71.5% 55.7% 61.0% 37.4% 46.0% 61.6% 84.0% Fish 52.1% 62.2% 70.6% 74.3% 57.2% 62.7% 33.2% 40.9% 56.9% 77.6% Bread 48.1% 57.5% 42.0% 44.2% 41.9% 45.9% 43.0% 52.9% 62.2% 84.8% Fruits/ veg 64.4% 76.9% 68.1% 71.8% 72.3% 79.2% 54.7% 67.3% 64.5% 88.0% Note: “Unconditional”—among everyone, whether or not any benefits were redeemed; “Conditional”—among those who redeemed any benefits. Table 8 Effects of Distance to the Closest Supermarket on percentage Redeemed by WIC Category All WIC Sites Eastern Michigan Grand Rapids, Michigan Nevada Texas Unconditional Conditional Unconditional Conditional Unconditional Conditional Unconditional Conditional Unconditional Conditional Milk 0.002 0.001 0.007** 0.004 −0.029** −0.027** 0 0.001 −0.006 −0.005 Cheese −0.001 −0.001 0.004 0.001 −0.01 −0.006 −0.001 −0.001 −0.007 −0.006 Eggs −0.004* −0.006*** 0 −0.002 −0.013 −0.009 −0.005 −0.006** −0.004 −0.002 Juice 0.002 0.002 0.007* 0.004 −0.033** −0.030*** 0.001 0.002 −0.004 −0.002 Cereal 0.002 0.002 0.004 0.002 −0.033*** −0.031*** 0.002 0.003 −0.004 −0.001 Beans 0 −0.001 0.002 0 −0.003 0.001 −0.001 −0.001 −0.011 −0.011 Fish 0.002 0.001 0.005 0.002 −0.008 −0.004 0 0 −0.009 −0.008 Bread 0.003 0.003 0.007** 0.006* −0.006 −0.003 −0.001 0 −0.003 −0.001 Fruits/veg 0 −0.001 0 −0.002 −0.014 −0.011 0 0.001 −0.003 −0.001 All WIC Sites Eastern Michigan Grand Rapids, Michigan Nevada Texas Unconditional Conditional Unconditional Conditional Unconditional Conditional Unconditional Conditional Unconditional Conditional Milk 0.002 0.001 0.007** 0.004 −0.029** −0.027** 0 0.001 −0.006 −0.005 Cheese −0.001 −0.001 0.004 0.001 −0.01 −0.006 −0.001 −0.001 −0.007 −0.006 Eggs −0.004* −0.006*** 0 −0.002 −0.013 −0.009 −0.005 −0.006** −0.004 −0.002 Juice 0.002 0.002 0.007* 0.004 −0.033** −0.030*** 0.001 0.002 −0.004 −0.002 Cereal 0.002 0.002 0.004 0.002 −0.033*** −0.031*** 0.002 0.003 −0.004 −0.001 Beans 0 −0.001 0.002 0 −0.003 0.001 −0.001 −0.001 −0.011 −0.011 Fish 0.002 0.001 0.005 0.002 −0.008 −0.004 0 0 −0.009 −0.008 Bread 0.003 0.003 0.007** 0.006* −0.006 −0.003 −0.001 0 −0.003 −0.001 Fruits/veg 0 −0.001 0 −0.002 −0.014 −0.011 0 0.001 −0.003 −0.001 Note: Two-sided hypothesis test; asterisks *** = statistically significant, p<0.01; ** = statistically significant, p<0.05; * = statistically significant, p<0.10. Table 8 Effects of Distance to the Closest Supermarket on percentage Redeemed by WIC Category All WIC Sites Eastern Michigan Grand Rapids, Michigan Nevada Texas Unconditional Conditional Unconditional Conditional Unconditional Conditional Unconditional Conditional Unconditional Conditional Milk 0.002 0.001 0.007** 0.004 −0.029** −0.027** 0 0.001 −0.006 −0.005 Cheese −0.001 −0.001 0.004 0.001 −0.01 −0.006 −0.001 −0.001 −0.007 −0.006 Eggs −0.004* −0.006*** 0 −0.002 −0.013 −0.009 −0.005 −0.006** −0.004 −0.002 Juice 0.002 0.002 0.007* 0.004 −0.033** −0.030*** 0.001 0.002 −0.004 −0.002 Cereal 0.002 0.002 0.004 0.002 −0.033*** −0.031*** 0.002 0.003 −0.004 −0.001 Beans 0 −0.001 0.002 0 −0.003 0.001 −0.001 −0.001 −0.011 −0.011 Fish 0.002 0.001 0.005 0.002 −0.008 −0.004 0 0 −0.009 −0.008 Bread 0.003 0.003 0.007** 0.006* −0.006 −0.003 −0.001 0 −0.003 −0.001 Fruits/veg 0 −0.001 0 −0.002 −0.014 −0.011 0 0.001 −0.003 −0.001 All WIC Sites Eastern Michigan Grand Rapids, Michigan Nevada Texas Unconditional Conditional Unconditional Conditional Unconditional Conditional Unconditional Conditional Unconditional Conditional Milk 0.002 0.001 0.007** 0.004 −0.029** −0.027** 0 0.001 −0.006 −0.005 Cheese −0.001 −0.001 0.004 0.001 −0.01 −0.006 −0.001 −0.001 −0.007 −0.006 Eggs −0.004* −0.006*** 0 −0.002 −0.013 −0.009 −0.005 −0.006** −0.004 −0.002 Juice 0.002 0.002 0.007* 0.004 −0.033** −0.030*** 0.001 0.002 −0.004 −0.002 Cereal 0.002 0.002 0.004 0.002 −0.033*** −0.031*** 0.002 0.003 −0.004 −0.001 Beans 0 −0.001 0.002 0 −0.003 0.001 −0.001 −0.001 −0.011 −0.011 Fish 0.002 0.001 0.005 0.002 −0.008 −0.004 0 0 −0.009 −0.008 Bread 0.003 0.003 0.007** 0.006* −0.006 −0.003 −0.001 0 −0.003 −0.001 Fruits/veg 0 −0.001 0 −0.002 −0.014 −0.011 0 0.001 −0.003 −0.001 Note: Two-sided hypothesis test; asterisks *** = statistically significant, p<0.01; ** = statistically significant, p<0.05; * = statistically significant, p<0.10. We also explored whether some food categories are more sensitive to distance (e.g., perishables or high weight to value). Table 8 shows the change in proportion of benefits redeemed in each category for each additional mile between the household and the supermarket where they could redeem benefits, both modeling unconditional redemption (i.e., including those who redeem no benefits) and conditional redemption (i.e., excluding those who redeem no benefits). Here, as in the rest of the paper, we focus on the unconditional analysis. Again, considering all WIC sites together and each site separately, the farther the distance to the nearest supermarket, the lower the redemption rate for eggs benefits. In Grand Rapids, households that are farther away from their nearest supermarket or superstore redeem statistically significantly less milk, juice, and cereal (but not eggs). In Eastern Michigan, households that are farther away redeem statistically significantly more milk, bread, and juice. The lack of any relation in two of the sites and the unexpected sign in one of the sites lead us to infer that the observed relation is spurious, probably due to multiple comparisons issues (i.e., we are testing a lot of hypotheses, some of them will appear significant simply by chance). As with the site-specific results, these food category-specific results should be interpreted with caution. Perhaps distance matters for some food categories, but not for others. Alternatively, the results may be due to chance. Discussion Among U.S. nutrition assistance programs, redemption rates are much higher for SNAP benefits than for WIC benefits. Some prior work suggests that living farther away from supermarkets may create barriers to a family’s ability to access affordable healthy food. Work by Rose and Richards (2004) found that a greater distance to supermarkets was associated with significantly lower fruit consumption among SNAP participants. Other work suggests that opening a new supermarket in a food desert can lead to increased fruit and vegetable consumption for low-income families (Cummins, Flint, and Matthews 2014; Elbel et al. 2017). This paper provides three results on SEBTC redemption and distance to retailers, and more broadly on distance to retailers as an explanation of why redemption rates are lower for WIC than for SNAP. First, the relationship between distance and redemption rates is quite small. Our results imply that even if every SNAP retailer also accepted WIC, redemption rates for WIC-participating families would not meaningfully increase. Similarly, if every SEBTC-WIC household had a participating retailer next door, redemption rates would barely change. Second, our analysis of the distances households travel to redeem benefits does not support the hypothesis that the longer average distance to WIC retailers represents an important explanation of lower WIC redemption rates. Weighting by dollars redeemed, on average SEBTC-SNAP households redeem their benefits at retailers 6.3 times as far away as the closest participating retailer, and 2.7 times as far as the closest supermarket/superstore. For SEBTC-WIC households, the corresponding figures are slightly smaller: 3.0 times and 2.5 times, respectively. Third, disaggregating WIC redemption rates by broad food category suggests that, with the exception of eggs, for no food category is redemption strongly related to distance. In contrast, redemption varies strongly with food category. In-depth qualitative interviews with households and a survey might provide insight into the relation between WIC’s restrictive food choices and lower redemption rates. It is possible that participants are confused by the various restrictions, or the WIC packages may not be appropriately constructed to meet participant needs. However, our results are inconsistent with any role of proximity to a retailer, even for perishable items. Together these results suggest that distance to retailers was not a major consideration in benefit redemption for Summer EBT households in 2012. Given that these results suggest that households do not shop at the closest retailer or even supermarket/superstore, it seems unlikely that distance to closest retailer or supermarket/superstore represents an important causal factor for the overwhelming share of households. These results are broadly consistent with recent developments in the “food desert” literature suggesting that food desert correlations—that is, that nutritional outcomes are worse in food deserts—are not causal. However, our sample does not contain participants in more rural areas with longer distances to retailers, so our results are not generalizable to this population. Our analyses also only include households with school-age children. It is possible that redemption patterns and the sensitivity to distance is different for households without school-age children (e.g., households that only have younger children, households with no children, single-person households). Footnotes 1 We chose to focus on one year of data to account for the fact that retailer options and location change over time. In this case, a cross sectional study is most appropriate. The study year 2012 provided us the largest sample and the simplest study design. For example, in 2013, households received $60 or $30 in benefits, but no households received zero benefits. 2 While whether a site adopted SEBTC-SNAP or SEBTC-WIC was not randomly assigned, our analyses control for observed differences in households; see below. Nevertheless, the small number of sites means that the results should be interpreted with caution. 3 Retailers include all stores where participants can redeem benefits. In addition to supermarkets and superstores, this may also include specialty stores such as butchers or bakeries, wholesalers, and convenience stores. Because the majority of redemptions happen at supermarkets and superstores, we limit our main analyses to these categories. 4 Ten miles is the maximum distance between a household and a retailer. See appendix A for sample deletion decisions. References Allcott H. , Diamond R. , Dube J.P . 2015 . The Geography of Poverty and Nutrition: Food Deserts and Food Choices across the U.S. New York University Working Paper. Castner L. , Henke J . 2011 . Benefit Redemption Patterns in the Supplemental Nutrition Assistance Program (No. b746c9a56cb34547b475799386b0182a). Washington, D.C: Mathematica Policy Research. Collins A. , Briefel R. , Klerman J.A. , Rowe G. , Wolf A. , Logan C.W. , Gordon A. et al. 2013 . Summer Electronic Benefits Transfer for Children (SEBTC) Demonstration: 2012 Final Report. Nutrition Assistance Program Report. Washington DC: U.S. Department of Agriculture, Food and Nutrition Service, Office of Policy Support. Cummins S. , Flint E. , Matthews S.A . 2014 . New Neighborhood Grocery Store Increased Awareness of Food Access but Did Not Alter Dietary Habits or Obesity . Health Affairs 33 ( 2 ): 283 – 91 . Google Scholar CrossRef Search ADS PubMed Dubowitz T. , Ghosh-Dastidar M. , Cohen D.A. , Beckman R. , Steiner E.D. , Hunter G.P. , Florez K.R. , et al. 2015 . Diet and Perceptions Change with Supermarket Introduction in Food Desert, But Not Because of Supermarket Use . Health Affairs 34 ( 11 ): 1858 – 68 . Google Scholar CrossRef Search ADS PubMed Edin K. , Boyd M. , Mabli J. , Ohls J. , Worthington J. , Greene S. , Redel N. , et al. 2013 . SNAP Food Security In-depth Interview Study Final Report . Alexandria VA : U.S. Department of Agriculture, Food and Nutrition Service . Elbel B. , Mijanovich T. , Kiszko K. , Abrams C. , Cantor J. , Dixon L.B . 2017 . The Introduction of a Supermarket via Tax-Credits in a Low-Income Area: The Influence on Purchasing and Consumption . American Journal of Health Promotion 31 ( 1 ): 59 – 66 . Google Scholar CrossRef Search ADS PubMed Gray K.F. , Cunningham K . 2016 . Trends in Supplemental Nutrition Assistance Program Participation Rates: Fiscal Year 2010 to Fiscal Year 2014 . Washington, D.C : Mathematica Policy Research . Grindal T. , Wilde P. , Schwartz G. , Klerman J. , Bartlett S. , Berman D . 2016 . Does Food Retail Access Moderate the Impact of Fruit and Vegetable Incentives for SNAP Participants? Evidence from Western Massachusetts . Food Policy 61 : 59 – 69 . Google Scholar CrossRef Search ADS Handbury J. , Rahkovsky I. , Schnell M . 2016 . Is the Focus on Food Deserts Fruitless? Retail Access and Food Purchases across the Socioeconomic Spectrum. The Wharton School Research Paper No. 91. Available at: https://papers.ssrn.com/sol3/papers.cfm?abstract_id=2757763. Institute of Medicine and National Research Council . 2009 . The Public Health Effects of Food Deserts: Workshop Summary. Washington DC : The National Academies Press . Johnson P. , Giannarelli L. , Huber E. , Betson D . 2015 . National and State-Level Estimates of Special Supplemental Nutrition Program for Women, Infants, and Children (WIC) Eligibles and Program Reach, 2011 . Alexandria VA : U.S. Department of Agriculture Food and Nutrition Service . Kyureghian G. , Nayga R.M. , Bhattacharya S . 2012 . The Effect of Food Store Access and Income on Household Purchases of Fruits and Vegetables: A Mixed Effects Analysis . Applied Economics Policy and Perspectives 35 ( 1 ): 69 – 88 . Google Scholar CrossRef Search ADS 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 ): 74 – 81 . Google Scholar CrossRef Search ADS Mabli J. , Worthington J . 2015 . The Food Access Environment and Food Purchase Behavior of SNAP Households . Journal of Hunger and Environmental Nutrition 10 : 132 – 49 . Google Scholar CrossRef Search ADS Phillips D. , Bell L. , Morgan R. , Pooler J . 2014 . Review of Impact and Examination of Participant Redemption Patterns . Washington DC : U.S. Department of Agriculture . Rose D. , Richards R . 2004 . Food Store Access and Household Fruit and Vegetable Use among Participants in the U.S. Food Stamp Program . Public Health Nutrition 7 ( 8 ): 1081 – 8 . Google Scholar CrossRef Search ADS PubMed Schwartz G. , Grindal T. , Wilde P. , Klerman J. , Bartlett S . 2017 . Supermarket Shopping and The Food Retail Environment among SNAP Participants . Journal of Hunger & Environmental Nutrition 1 – 26 . U.S. Department of Agriculture, Economic Research Service . 2009 . Access to Affordable and Nutritious Food: Measuring and Understanding Food Deserts and Their Consequences. Report to Congress. Administrative Publication No. 36, June 2009. Available at: http://ers.usda.gov/publications/ap-administrative-publication/ap-036.aspx. Castner L. , Henke J . 2011 . Benefit Redemption Patterns in the Supplemental Nutrition Assistance Program . Alexandria VA : U.S. Department of Agriculture Food and Nutrition Service, Office of Research and Analysis . U.S. Department of Agriculture, Food and Nutrition Service . 2016a . Benefit Redemption Patterns in the Supplemental Nutrition Assistance Program. Available at: https://www.fns.usda.gov/snap/benefit-redemption-patterns-supplemental-nutrition-assistance-program. U.S. Department of Agriculture, Food and Nutrition Service . 2016b . WIC Program. Available at: https://www.fns.usda.gov/pd/wic-program. U.S. Department of Agriculture, Food and Nutrition Service . 2017a . Women, Infants and Children (WIC): WIC Eligibility Requirements. Available at: https://www.fns.usda.gov/wic/wic-eligibility-requirements. U.S. Department of Agriculture, Food and Nutrition Service . 2017b . Supplemental Nutrition Assistance Program (SNAP): Eligibility. Available at: https://www.fns.usda.gov/snap/eligibility. Ver Ploeg M. , Mancino L. , Todd J.E. , Clay D.M. , Scharadin B . 2015 . Where Do Americans Usually Shop for Food and How Do They Travel to Get there? Initial Findings from the National Household Food Acquisition and Purchase Survey . Washington DC : U.S. Department of Agriculture, Economic Research Service . ver Ploeg S. , Wilde P . Forthcoming. How Do Food Retail Choices Vary Within and Between Food Retail Environments? Food Policy . Appendix A: Data File Construction This appendix discusses our data sources, geocoding, and sample deletions. A.1 Data Sources Our data come from several sources: SEBTC Sample files, SEBTC EBT records, FNS Retailer listings and the American Community Survey from the U.S. Census Bureau. Each data source and how it is used is described below. Household Data: Our data came from several sources. SEBTC Study Sample data, initially created to verify eligibility for SEBTC and to support stratified random assignment, provided randomization status (i.e., treatment/control), address (for geocoding), and household characteristics (as in table 3). SEBTC EBT data provided information on benefits redeemed. We follow the main SEBTC analysis in computing redemption rates (e.g., Collins et al. 2013). Grantees using the WIC model also provided data on the date, time, and dollar value of each transaction. The data for these sites permit the analysis of redemptions at the food category level. The original issuance data did not include the dollar value of benefits issued. To aggregate across food category levels (e.g., overall redemption rates), the average cost per unit for each food category was imputed, based upon the redemption data. It is thus possible for a household to redeem more than 100% of the benefit (i.e., if costs per unit were higher than average). In practice, such deviations are rare and small. Retailer Data: USDA/FNS provided administrative data on SNAP retailers for each participating state: exact address, store type (which we recoded as supermarket/superstore or not), and participates in WIC (yes/no). We do not have retailer data for neighboring states. Thus, our analysis involves distance to the closest retailer where benefits could be redeemed within the same state. In some cases, the closest retailer to a household may be across the state border, and our analysis cannot account for that. Our main analysis drops households within 1 mile of a state border. Appendix Section A.4 provides sensitivity analyses, including households living within 1 mile of the border. Census Tract Characteristics: To further control for household differences—especially those that might be correlated with distance to a retailer—we augmented the quite limited case-level information with census tract-level information from the American Community Survey (ACS), conducted by the U.S. Census. Geocoding (discussed below) provided census tract for each SEBTC household. ACS samples in any year are too small to generate reliable tract-level estimates. Instead, reported tract-level estimates are centered five-year estimates. Because the study occurred in the summer of 2012, we pulled data from the American Community Survey 2012 5-Year Estimates (available at: https://factfinder.census.gov/faces/nav/jsf/pages/index.xhtml). Population density estimates were drawn from the 2010 Census (available at: https://www.census.gov/quickfacts/meta/long_POP060210.htm and https://www.census.gov/geo/maps-data/data/tiger.html). See table 3 for the specific tract level variables used. A.2 Geocoding We used ArcGIS Online Geocoding Service to geocode (i.e., assign longitude and latitude) to each household and retailer. We then used the Network Analyst tools within ArcGIS to compute the distance from each household to the closest retailer in each class: SNAP/WIC and any/supermarket or superstore. A.3 Sample Deletions Table A.1 describes sample deletions, overall and by site. In 2012, 34,210 households were assigned to the SEBTC treatment group. From that count to the analysis file, we made the following deletions: Ungeocodable: This is an analysis of the impact of distance, where distance is computed via geocoding. We therefore had to drop any records that were not geocoding: no address, P.O. Box, apparently proper address that was not geocodable (see discussion in the next section), distance could not be computed (geocoding software could not connect the address to the road network), and address not in state granting benefits. Tribal Sites: The tribal sites had a high rate of ungeocodable addresses (approximately 12% of all households in tribal sites). The tribal sites were dropped from the main analysis because they may not be representative, considering the high rate of missing addresses. Outliers. Almost all of the households not dropped as ungeocodable were within 10 miles of a retailer (well over 99 percent; see figure A.1). So as not to have outliers drive the regression results, we dropped all households with a distance of greater than 10 miles to closest retailer where they could redeem benefits. Missing Outcomes. Some households could not be matched to EBT records in at least one month. We dropped any household for which we did not have EBT data for every month sufficient to construct the four outcomes considered in this study: redeemed any benefits, redeemed any benefits in a month, percentage of benefits redeemed (unconditional on redeeming any benefits), redeemed all benefits (defined in two ways: as redeeming the maximum amount of benefits in any single month of the demonstration (ever exhausted), and exhausting all benefits over the course of the demonstration (exhausting all)). Appendix B: Sensitivity to Analytic Choices and Additional Substantive Results This appendix provides a sensitivity analysis for analytic choices (Section B.1) and substantive results for additional outcomes (Section B.2). This section also considers sensitivity of the results to sample deletion choices (Section B.3). B.1 Sensitivity to Analytic Choices This section considers sensitivity of the results to analytic choices. Table B.1.1 shows how the key regression results vary with the choice of covariates. Model 1 contains no additional regressors, only the distance measure. Model 2 includes indicators for each of the 14 sites included in the analyses. Model 3 includes the covariates from the sample file in addition to the site-level indicators. Model 4 includes the sample covariates, the site-level indicators, and the census tract characteristics from the American Community Survey. When no additional regressors are included, there is a significant, negative relationship between the proportion of benefits redeemed and the distance to the nearest supermarket where benefits could be redeemed. For each additional mile farther between a household and a supermarket, a participant would redeem 2% less of his or her benefits. A household that was a maximum of 10 miles away would redeem 20% less of its benefits. For SNAP households alone, the relationship is much smaller, and positive. The relationship seems to be driven by the WIC sites. However, as soon as we control for any additional covariates, the relationship disappears. The relationship between distance and redemption rates appears to be spurious. We ran several variations of our analytic model on the focal outcome (unconditional percentage of benefits redeemed), and additional model specifications yielded the same results. First, we looked at the relationship between distance to the nearest retailer of any type where a participant could redeem benefits (as opposed to closest supermarket). There seems to be no relationship between distance to the closest retailer where participants could redeem benefits and the proportion of their benefits they actually redeem. See table B.1.2. We also examined the relationship between distance to the nearest supermarket where participants could redeem benefits, and redemption patterns, controlling for the distance to the nearest store of any type where participants could redeem benefits. Results are shown in table B.1.3. The pooled results show no relationship between distance to either store type or proportion of benefits redeemed. In SNAP sites, when controlling for distance to the store of any type, there was a small, negative relationship between distance to the nearest supermarket and proportion of benefits redeemed, although the relationship is insignificant and quite close to 0. Lastly, we controlled for a non-linear relationship between distance and redemption patterns, shown in table B.1.4. Controlling for a spline at 1 mile, there is no significant relationship between distance to the nearest supermarket and redemption rates. However, in SNAP sites, a participant who is more than 10 miles away from the nearest retailer would redeem approximately 10% more of his or her benefits. The pattern does not hold when the spline is at ¾ of a mile (table B.1.4) B.2 Substantive Results for Additional Outcomes We also conducted analyses on three additional outcomes: whether a household redeemed any benefit, whether a household exhausted its benefits in any month of the SEBTC demonstration, and whether a household exhausted all of its benefits over the course of the demonstration. Table B.2.1 shows the results for participants redeeming any of their benefits, proportion redeemed (conditional on redeeming any benefit), exhausted benefits in any month, and exhausted benefits in all months. These models include the full set of covariates. The patterns for redeeming any benefit are similar to the patterns for proportion of benefits redeemed. There is no detectable relationship between distance and redemption, exhausting benefits in a given month, or proportion of benefits redeemed (conditional on redeeming any benefit). However, for SNAP households, there is a negative, significant relationship between distance to the nearest supermarket where participants could redeem benefits and their likelihood of exhausting all of their benefits over the course of the SEBTC demonstration. A household that is 10 miles farther from a supermarket would be 7% less likely to redeem all of its benefits. B.3 Sensitivity to Sample Selection This section considers sensitivity of the results to sample deletion choices. The analysis is impossible without addresses, so an analysis of the impact of losing records to failure to geocode is not possible. Table B.3.1 presents the main results for: (a) the analysis sample; (b) adding those within 1 mile of a state border; and (c) adding the two tribal sites (but still excluding those within 1 mile of a state border). The results are largely consistent for each sample. When the cases close to state borders are included, there are no changes to the coefficients. When the tribal sites are included, the coefficient for WIC is slightly larger, but still insignificant. Table A.1.1 Sample Attrition by Site Total Households Missing Address PO Box Could Not Be Geocoded No Distance Calculation Out of State Outlier Distance (>10 miles) < 1 Mile to State Border Missing EBT Data Analysis Sample Total Households 34,520 1 782 11 45 10 115 733 2,524 30,299 All SNAP Sites 17,738 1 98 9 31 1 11 727 283 16,577 Eastern Connecticut 2,276 1 11 0 0 1 6 23 70 2,164 Western Connecticut 1,286 0 1 0 0 0 0 0 28 1,257 Delaware 2,870 0 40 0 9 0 0 0 3 2,818 Kansas City, MO 2,970 0 0 1 1 0 0 123 79 2,766 St. Louis, MO 3,468 0 1 0 20 0 0 488 9 2,950 Deschutes, OR 1,728 0 7 8 1 0 5 0 90 1,617 Salem, OR 1,573 0 14 0 0 0 0 0 0 1,559 Washington 1,567 0 24 0 0 0 0 93 4 1,446 All WIC Sites 16,782 0 684 2 14 9 104 6 2,241 13,722 Cherokee 2,748 0 589 0 1 5 18 3 375 1,757 Chickasaw 2,559 0 63 1 0 0 58 1 464 1,972 Eastern Michigan 2,630 0 24 0 1 0 0 0 1,183 1,422 Grand Rapids, MI 2,782 0 1 1 0 0 0 0 7 2,773 Nevada 3,049 0 7 0 12 4 25 2 112 2,887 Texas 3,014 0 0 0 0 0 3 0 100 2,911 Total Households Missing Address PO Box Could Not Be Geocoded No Distance Calculation Out of State Outlier Distance (>10 miles) < 1 Mile to State Border Missing EBT Data Analysis Sample Total Households 34,520 1 782 11 45 10 115 733 2,524 30,299 All SNAP Sites 17,738 1 98 9 31 1 11 727 283 16,577 Eastern Connecticut 2,276 1 11 0 0 1 6 23 70 2,164 Western Connecticut 1,286 0 1 0 0 0 0 0 28 1,257 Delaware 2,870 0 40 0 9 0 0 0 3 2,818 Kansas City, MO 2,970 0 0 1 1 0 0 123 79 2,766 St. Louis, MO 3,468 0 1 0 20 0 0 488 9 2,950 Deschutes, OR 1,728 0 7 8 1 0 5 0 90 1,617 Salem, OR 1,573 0 14 0 0 0 0 0 0 1,559 Washington 1,567 0 24 0 0 0 0 93 4 1,446 All WIC Sites 16,782 0 684 2 14 9 104 6 2,241 13,722 Cherokee 2,748 0 589 0 1 5 18 3 375 1,757 Chickasaw 2,559 0 63 1 0 0 58 1 464 1,972 Eastern Michigan 2,630 0 24 0 1 0 0 0 1,183 1,422 Grand Rapids, MI 2,782 0 1 1 0 0 0 0 7 2,773 Nevada 3,049 0 7 0 12 4 25 2 112 2,887 Texas 3,014 0 0 0 0 0 3 0 100 2,911 Table A.1.1 Sample Attrition by Site Total Households Missing Address PO Box Could Not Be Geocoded No Distance Calculation Out of State Outlier Distance (>10 miles) < 1 Mile to State Border Missing EBT Data Analysis Sample Total Households 34,520 1 782 11 45 10 115 733 2,524 30,299 All SNAP Sites 17,738 1 98 9 31 1 11 727 283 16,577 Eastern Connecticut 2,276 1 11 0 0 1 6 23 70 2,164 Western Connecticut 1,286 0 1 0 0 0 0 0 28 1,257 Delaware 2,870 0 40 0 9 0 0 0 3 2,818 Kansas City, MO 2,970 0 0 1 1 0 0 123 79 2,766 St. Louis, MO 3,468 0 1 0 20 0 0 488 9 2,950 Deschutes, OR 1,728 0 7 8 1 0 5 0 90 1,617 Salem, OR 1,573 0 14 0 0 0 0 0 0 1,559 Washington 1,567 0 24 0 0 0 0 93 4 1,446 All WIC Sites 16,782 0 684 2 14 9 104 6 2,241 13,722 Cherokee 2,748 0 589 0 1 5 18 3 375 1,757 Chickasaw 2,559 0 63 1 0 0 58 1 464 1,972 Eastern Michigan 2,630 0 24 0 1 0 0 0 1,183 1,422 Grand Rapids, MI 2,782 0 1 1 0 0 0 0 7 2,773 Nevada 3,049 0 7 0 12 4 25 2 112 2,887 Texas 3,014 0 0 0 0 0 3 0 100 2,911 Total Households Missing Address PO Box Could Not Be Geocoded No Distance Calculation Out of State Outlier Distance (>10 miles) < 1 Mile to State Border Missing EBT Data Analysis Sample Total Households 34,520 1 782 11 45 10 115 733 2,524 30,299 All SNAP Sites 17,738 1 98 9 31 1 11 727 283 16,577 Eastern Connecticut 2,276 1 11 0 0 1 6 23 70 2,164 Western Connecticut 1,286 0 1 0 0 0 0 0 28 1,257 Delaware 2,870 0 40 0 9 0 0 0 3 2,818 Kansas City, MO 2,970 0 0 1 1 0 0 123 79 2,766 St. Louis, MO 3,468 0 1 0 20 0 0 488 9 2,950 Deschutes, OR 1,728 0 7 8 1 0 5 0 90 1,617 Salem, OR 1,573 0 14 0 0 0 0 0 0 1,559 Washington 1,567 0 24 0 0 0 0 93 4 1,446 All WIC Sites 16,782 0 684 2 14 9 104 6 2,241 13,722 Cherokee 2,748 0 589 0 1 5 18 3 375 1,757 Chickasaw 2,559 0 63 1 0 0 58 1 464 1,972 Eastern Michigan 2,630 0 24 0 1 0 0 0 1,183 1,422 Grand Rapids, MI 2,782 0 1 1 0 0 0 0 7 2,773 Nevada 3,049 0 7 0 12 4 25 2 112 2,887 Texas 3,014 0 0 0 0 0 3 0 100 2,911 Figure A.1 View largeDownload slide Percent of benefits redeemed by distance to the nearest supermarket of superstore Figure A.1 View largeDownload slide Percent of benefits redeemed by distance to the nearest supermarket of superstore Table B.1.1 Effects of Distance to the Closest Supermarket on Percentage Redeemed for Multiple Regression Models Model 1 Model 2 Model 3 Model 4 SNAP sites, 8 sites Driving distance to the nearest supermarket where benefits could be redeemed 0.002* −0.000 −0.000 −0.001 (0.001) (0.001) (0.001) (0.001) Constant 0.936*** 0.959*** 0.909*** 0.878*** (0.002) (0.004) (0.060) (0.066) R-squared 0.000 0.016 0.025 0.026 N 16,577 16,577 16,577 16,577 WIC sites, 4 sites Driving distance to the nearest supermarket where benefits could be redeemed −0.017*** 0.001 0.001 0.001 (0.002) (0.002) (0.002) (0.002) Constant 0.604*** 0.641*** 0.526*** 0.680*** (0.004) (0.008) (0.061) (0.074) R-squared 0.007 0.052 0.070 0.075 N 9,993 9,993 9,993 9,993 Model 1 Model 2 Model 3 Model 4 SNAP sites, 8 sites Driving distance to the nearest supermarket where benefits could be redeemed 0.002* −0.000 −0.000 −0.001 (0.001) (0.001) (0.001) (0.001) Constant 0.936*** 0.959*** 0.909*** 0.878*** (0.002) (0.004) (0.060) (0.066) R-squared 0.000 0.016 0.025 0.026 N 16,577 16,577 16,577 16,577 WIC sites, 4 sites Driving distance to the nearest supermarket where benefits could be redeemed −0.017*** 0.001 0.001 0.001 (0.002) (0.002) (0.002) (0.002) Constant 0.604*** 0.641*** 0.526*** 0.680*** (0.004) (0.008) (0.061) (0.074) R-squared 0.007 0.052 0.070 0.075 N 9,993 9,993 9,993 9,993 Note: Model 1: no regressors; Model 2: site indicators only; Model 3: site indicators and household characteristics; Model 4: site indicators, household characteristics for the household itself and household characteristics for the census tract (computed from 5-year ACS tabulations). Two-sided hypothesis test; asterisks *** = statistically significant, p<0.01; ** = statistically significant, p<0.05; * = statistically significant, p<0.10. View Large Table B.1.1 Effects of Distance to the Closest Supermarket on Percentage Redeemed for Multiple Regression Models Model 1 Model 2 Model 3 Model 4 SNAP sites, 8 sites Driving distance to the nearest supermarket where benefits could be redeemed 0.002* −0.000 −0.000 −0.001 (0.001) (0.001) (0.001) (0.001) Constant 0.936*** 0.959*** 0.909*** 0.878*** (0.002) (0.004) (0.060) (0.066) R-squared 0.000 0.016 0.025 0.026 N 16,577 16,577 16,577 16,577 WIC sites, 4 sites Driving distance to the nearest supermarket where benefits could be redeemed −0.017*** 0.001 0.001 0.001 (0.002) (0.002) (0.002) (0.002) Constant 0.604*** 0.641*** 0.526*** 0.680*** (0.004) (0.008) (0.061) (0.074) R-squared 0.007 0.052 0.070 0.075 N 9,993 9,993 9,993 9,993 Model 1 Model 2 Model 3 Model 4 SNAP sites, 8 sites Driving distance to the nearest supermarket where benefits could be redeemed 0.002* −0.000 −0.000 −0.001 (0.001) (0.001) (0.001) (0.001) Constant 0.936*** 0.959*** 0.909*** 0.878*** (0.002) (0.004) (0.060) (0.066) R-squared 0.000 0.016 0.025 0.026 N 16,577 16,577 16,577 16,577 WIC sites, 4 sites Driving distance to the nearest supermarket where benefits could be redeemed −0.017*** 0.001 0.001 0.001 (0.002) (0.002) (0.002) (0.002) Constant 0.604*** 0.641*** 0.526*** 0.680*** (0.004) (0.008) (0.061) (0.074) R-squared 0.007 0.052 0.070 0.075 N 9,993 9,993 9,993 9,993 Note: Model 1: no regressors; Model 2: site indicators only; Model 3: site indicators and household characteristics; Model 4: site indicators, household characteristics for the household itself and household characteristics for the census tract (computed from 5-year ACS tabulations). Two-sided hypothesis test; asterisks *** = statistically significant, p<0.01; ** = statistically significant, p<0.05; * = statistically significant, p<0.10. View Large Table B.1.2 Effects of Distance to the Closest Retailer on Percentage Redeemed SNAP Households WIC Households Driving distance to the nearest retailer where benefits could be redeemed 0.000 −0.000 (0.002) (0.002) Constant 0.877*** 0.683*** (0.066) (0.074) R-squared 0.026 0.075 N 16,577 9,993 SNAP Households WIC Households Driving distance to the nearest retailer where benefits could be redeemed 0.000 −0.000 (0.002) (0.002) Constant 0.877*** 0.683*** (0.066) (0.074) R-squared 0.026 0.075 N 16,577 9,993 Note: Two-sided hypothesis test; asterisks *** = statistically significant, p<0.01; ** = statistically significant, p<0.05; * = statistically significant, p<0.10. View Large Table B.1.2 Effects of Distance to the Closest Retailer on Percentage Redeemed SNAP Households WIC Households Driving distance to the nearest retailer where benefits could be redeemed 0.000 −0.000 (0.002) (0.002) Constant 0.877*** 0.683*** (0.066) (0.074) R-squared 0.026 0.075 N 16,577 9,993 SNAP Households WIC Households Driving distance to the nearest retailer where benefits could be redeemed 0.000 −0.000 (0.002) (0.002) Constant 0.877*** 0.683*** (0.066) (0.074) R-squared 0.026 0.075 N 16,577 9,993 Note: Two-sided hypothesis test; asterisks *** = statistically significant, p<0.01; ** = statistically significant, p<0.05; * = statistically significant, p<0.10. View Large Table B.1.3 Effects of Multiple Distances on Percentage Redeemed All Households SNAP Households WIC Households Driving distance to the nearest supermarket where benefits could be redeemed 0.000 −0.002 0.004 (0.002) (0.002) (0.003) Driving distance to the nearest retailer of any type 0.000 0.002 −0.004 (0.002) (0.003) (0.004) Constant 0.605*** 0.878*** 0.680*** (0.050) (0.066) (0.074) R-squared 0.345 0.026 0.075 N 26,570 16,577 9,993 All Households SNAP Households WIC Households Driving distance to the nearest supermarket where benefits could be redeemed 0.000 −0.002 0.004 (0.002) (0.002) (0.003) Driving distance to the nearest retailer of any type 0.000 0.002 −0.004 (0.002) (0.003) (0.004) Constant 0.605*** 0.878*** 0.680*** (0.050) (0.066) (0.074) R-squared 0.345 0.026 0.075 N 26,570 16,577 9,993 Note: Two-sided hypothesis test; asterisks *** = statistically significant, p<0.01; ** = statistically significant, p<0.05; * = statistically significant, p<0.10. View Large Table B.1.3 Effects of Multiple Distances on Percentage Redeemed All Households SNAP Households WIC Households Driving distance to the nearest supermarket where benefits could be redeemed 0.000 −0.002 0.004 (0.002) (0.002) (0.003) Driving distance to the nearest retailer of any type 0.000 0.002 −0.004 (0.002) (0.003) (0.004) Constant 0.605*** 0.878*** 0.680*** (0.050) (0.066) (0.074) R-squared 0.345 0.026 0.075 N 26,570 16,577 9,993 All Households SNAP Households WIC Households Driving distance to the nearest supermarket where benefits could be redeemed 0.000 −0.002 0.004 (0.002) (0.002) (0.003) Driving distance to the nearest retailer of any type 0.000 0.002 −0.004 (0.002) (0.003) (0.004) Constant 0.605*** 0.878*** 0.680*** (0.050) (0.066) (0.074) R-squared 0.345 0.026 0.075 N 26,570 16,577 9,993 Note: Two-sided hypothesis test; asterisks *** = statistically significant, p<0.01; ** = statistically significant, p<0.05; * = statistically significant, p<0.10. View Large Table B.1.4 Effects of Distance to the Closest Supermarket on Percentage Redeemed with Spline at 3/4 Mile SNAP Households WIC Households 0.75 Miles 1.0 Mile 0.75 Miles 1.0 Mile Driving distance to the nearest supermarket where benefits could be redeemed −0.000 0.000 0.001 0.000 (0.001) (0.001) (0.002) (0.002) Spline = 1 Mile 0.007 0.009* 0.007 −0.005 (0.007) (0.005) (0.015) (0.011) Constant 0.875*** 0.871*** 0.678*** 0.682*** (0.066) (0.066) (0.075) (0.075) R-squared 0.026 0.026 0.075 0.075 N 16,577 16,577 9,993 9,993 SNAP Households WIC Households 0.75 Miles 1.0 Mile 0.75 Miles 1.0 Mile Driving distance to the nearest supermarket where benefits could be redeemed −0.000 0.000 0.001 0.000 (0.001) (0.001) (0.002) (0.002) Spline = 1 Mile 0.007 0.009* 0.007 −0.005 (0.007) (0.005) (0.015) (0.011) Constant 0.875*** 0.871*** 0.678*** 0.682*** (0.066) (0.066) (0.075) (0.075) R-squared 0.026 0.026 0.075 0.075 N 16,577 16,577 9,993 9,993 Note: Two-sided hypothesis test; asterisks *** = statistically significant, p<0.01; ** = statistically significant, p<0.05; * = statistically significant, p<0.10. View Large Table B.1.4 Effects of Distance to the Closest Supermarket on Percentage Redeemed with Spline at 3/4 Mile SNAP Households WIC Households 0.75 Miles 1.0 Mile 0.75 Miles 1.0 Mile Driving distance to the nearest supermarket where benefits could be redeemed −0.000 0.000 0.001 0.000 (0.001) (0.001) (0.002) (0.002) Spline = 1 Mile 0.007 0.009* 0.007 −0.005 (0.007) (0.005) (0.015) (0.011) Constant 0.875*** 0.871*** 0.678*** 0.682*** (0.066) (0.066) (0.075) (0.075) R-squared 0.026 0.026 0.075 0.075 N 16,577 16,577 9,993 9,993 SNAP Households WIC Households 0.75 Miles 1.0 Mile 0.75 Miles 1.0 Mile Driving distance to the nearest supermarket where benefits could be redeemed −0.000 0.000 0.001 0.000 (0.001) (0.001) (0.002) (0.002) Spline = 1 Mile 0.007 0.009* 0.007 −0.005 (0.007) (0.005) (0.015) (0.011) Constant 0.875*** 0.871*** 0.678*** 0.682*** (0.066) (0.066) (0.075) (0.075) R-squared 0.026 0.026 0.075 0.075 N 16,577 16,577 9,993 9,993 Note: Two-sided hypothesis test; asterisks *** = statistically significant, p<0.01; ** = statistically significant, p<0.05; * = statistically significant, p<0.10. View Large Table B.2.1 Regression Results for Alternate Outcomes Redeemed Any Benefit Conditional Proportion Redeemed Exhausted Benefits in Any Month Exhausted Benefits in All Months SNAP sites, 8 sites Driving distance to the nearest supermarket where benefits could be redeemed −0.001 −0.000 −0.003 −0.007** (0.001) (0.001) (0.003) (0.003) Constant 0.950*** 0.924*** 0.854*** −0.007 (0.057) (0.042) (0.077) (0.096) R-squared 0.016 0.016 0.067 0.127 N 16,577 15,923 16,577 16,577 WIC sites, 8 sites Driving distance to the nearest supermarket where benefits could be redeemed 0.001 0.000 0.001 0.000 (0.002) (0.001) (0.001) (0.000) Constant 0.654*** 0.997*** 0.551*** 0.024 (0.086) (0.047) (0.074) (0.017) R-squared 0.063 0.314 0.172 0.014 N 9,993 8,359 9,993 9,993 Redeemed Any Benefit Conditional Proportion Redeemed Exhausted Benefits in Any Month Exhausted Benefits in All Months SNAP sites, 8 sites Driving distance to the nearest supermarket where benefits could be redeemed −0.001 −0.000 −0.003 −0.007** (0.001) (0.001) (0.003) (0.003) Constant 0.950*** 0.924*** 0.854*** −0.007 (0.057) (0.042) (0.077) (0.096) R-squared 0.016 0.016 0.067 0.127 N 16,577 15,923 16,577 16,577 WIC sites, 8 sites Driving distance to the nearest supermarket where benefits could be redeemed 0.001 0.000 0.001 0.000 (0.002) (0.001) (0.001) (0.000) Constant 0.654*** 0.997*** 0.551*** 0.024 (0.086) (0.047) (0.074) (0.017) R-squared 0.063 0.314 0.172 0.014 N 9,993 8,359 9,993 9,993 Note: Two-sided hypothesis test; asterisks *** = statistically significant, p<0.01; ** = statistically significant, p<0.05; * = statistically significant, p<0.10. View Large Table B.2.1 Regression Results for Alternate Outcomes Redeemed Any Benefit Conditional Proportion Redeemed Exhausted Benefits in Any Month Exhausted Benefits in All Months SNAP sites, 8 sites Driving distance to the nearest supermarket where benefits could be redeemed −0.001 −0.000 −0.003 −0.007** (0.001) (0.001) (0.003) (0.003) Constant 0.950*** 0.924*** 0.854*** −0.007 (0.057) (0.042) (0.077) (0.096) R-squared 0.016 0.016 0.067 0.127 N 16,577 15,923 16,577 16,577 WIC sites, 8 sites Driving distance to the nearest supermarket where benefits could be redeemed 0.001 0.000 0.001 0.000 (0.002) (0.001) (0.001) (0.000) Constant 0.654*** 0.997*** 0.551*** 0.024 (0.086) (0.047) (0.074) (0.017) R-squared 0.063 0.314 0.172 0.014 N 9,993 8,359 9,993 9,993 Redeemed Any Benefit Conditional Proportion Redeemed Exhausted Benefits in Any Month Exhausted Benefits in All Months SNAP sites, 8 sites Driving distance to the nearest supermarket where benefits could be redeemed −0.001 −0.000 −0.003 −0.007** (0.001) (0.001) (0.003) (0.003) Constant 0.950*** 0.924*** 0.854*** −0.007 (0.057) (0.042) (0.077) (0.096) R-squared 0.016 0.016 0.067 0.127 N 16,577 15,923 16,577 16,577 WIC sites, 8 sites Driving distance to the nearest supermarket where benefits could be redeemed 0.001 0.000 0.001 0.000 (0.002) (0.001) (0.001) (0.000) Constant 0.654*** 0.997*** 0.551*** 0.024 (0.086) (0.047) (0.074) (0.017) R-squared 0.063 0.314 0.172 0.014 N 9,993 8,359 9,993 9,993 Note: Two-sided hypothesis test; asterisks *** = statistically significant, p<0.01; ** = statistically significant, p<0.05; * = statistically significant, p<0.10. View Large Table B.3.1 Sensitivity to Sample Deletion Analytic Model Including Household Near State Borders Including Tribal Sites SNAP sites, 8 sites Driving distance to the nearest supermarket where benefits could be redeemed −0.001 −0.000 −0.001 (0.001) (0.000) (0.001) Constant 0.878*** 0.876*** 0.878*** (0.066) (0.021) (0.066) R-squared 0.026 0.027 0.026 N 16,577 17,304 16,577 WIC sites, 4 sites Driving distance to the nearest supermarket where benefits could be redeemed 0.001 0.001 0.003 (0.002) (0.001) (0.001) Constant 0.680*** 0.560*** 0.546*** (0.074) (0.076) (0.071) R-squared 0.075 0.075 0.083 N 9,993 10,006 13,772 Analytic Model Including Household Near State Borders Including Tribal Sites SNAP sites, 8 sites Driving distance to the nearest supermarket where benefits could be redeemed −0.001 −0.000 −0.001 (0.001) (0.000) (0.001) Constant 0.878*** 0.876*** 0.878*** (0.066) (0.021) (0.066) R-squared 0.026 0.027 0.026 N 16,577 17,304 16,577 WIC sites, 4 sites Driving distance to the nearest supermarket where benefits could be redeemed 0.001 0.001 0.003 (0.002) (0.001) (0.001) Constant 0.680*** 0.560*** 0.546*** (0.074) (0.076) (0.071) R-squared 0.075 0.075 0.083 N 9,993 10,006 13,772 Note: Two-sided hypothesis test; asterisks *** = statistically significant, p<0.01; ** = statistically significant, p<0.05; * = statistically significant, p<0.10. View Large Table B.3.1 Sensitivity to Sample Deletion Analytic Model Including Household Near State Borders Including Tribal Sites SNAP sites, 8 sites Driving distance to the nearest supermarket where benefits could be redeemed −0.001 −0.000 −0.001 (0.001) (0.000) (0.001) Constant 0.878*** 0.876*** 0.878*** (0.066) (0.021) (0.066) R-squared 0.026 0.027 0.026 N 16,577 17,304 16,577 WIC sites, 4 sites Driving distance to the nearest supermarket where benefits could be redeemed 0.001 0.001 0.003 (0.002) (0.001) (0.001) Constant 0.680*** 0.560*** 0.546*** (0.074) (0.076) (0.071) R-squared 0.075 0.075 0.083 N 9,993 10,006 13,772 Analytic Model Including Household Near State Borders Including Tribal Sites SNAP sites, 8 sites Driving distance to the nearest supermarket where benefits could be redeemed −0.001 −0.000 −0.001 (0.001) (0.000) (0.001) Constant 0.878*** 0.876*** 0.878*** (0.066) (0.021) (0.066) R-squared 0.026 0.027 0.026 N 16,577 17,304 16,577 WIC sites, 4 sites Driving distance to the nearest supermarket where benefits could be redeemed 0.001 0.001 0.003 (0.002) (0.001) (0.001) Constant 0.680*** 0.560*** 0.546*** (0.074) (0.076) (0.071) R-squared 0.075 0.075 0.083 N 9,993 10,006 13,772 Note: Two-sided hypothesis test; asterisks *** = statistically significant, p<0.01; ** = statistically significant, p<0.05; * = statistically significant, p<0.10. View Large © The Author(s) 2018. Published by Oxford University Press on behalf of the Agricultural and Applied Economics Association. All rights reserved. For Permissions, please e-mail: journals.permissions@oup.com http://www.deepdyve.com/assets/images/DeepDyve-Logo-lg.png Applied Economic Perspectives and Policy Oxford University Press

Retailer Proximity and Nutrition Program Redemptions: Evidence From the Summer Electronic Benefit Transfer For Children Program

Loading next page...
 
/lp/ou_press/retailer-proximity-and-nutrition-program-redemptions-evidence-from-the-BEWeKJQFQE
Publisher
Oxford University Press
Copyright
© The Author(s) 2018. Published by Oxford University Press on behalf of the Agricultural and Applied Economics Association. All rights reserved. For Permissions, please e-mail: journals.permissions@oup.com
ISSN
2040-5790
eISSN
2040-5804
D.O.I.
10.1093/aepp/ppy003
Publisher site
See Article on Publisher Site

Abstract

Abstract Although nearly all Supplemental Nutrition Assistance Program benefits are redeemed, a moderate share of Special Supplemental Nutrition Program for Women, Infant, and Children (WIC) benefits go unredeemed. Some hypothesize that the redemption rate differences are due to the lower density of WIC-authorized retailers. For the 2012 Summer Electronic Benefit Transfer for Children sites, this paper finds no consistent evidence of a relationship between redemption rates and retailer proximity. In fact, households often travel past the closest participating retailer to redeem their benefits. These findings are consistent with recent “food deserts” literature, which suggests that correlations between retailer environment and nutritional outcomes are not causal. Rather, it appears that differences in redemption rates may be related to the restrictions on what foods can be redeemed in what form with WIC. SNAP, WIC, benefit redemption, food access, geospatial analysis Redemption rates for the Supplemental Nutrition Assistance Program (SNAP) are much higher than for the Special Nutrition Assistance Program for Women Infants and Children (WIC) (Castner and Henke 2011). Fewer retailers participate in WIC than in SNAP. It is thus plausible that differences in households’ geographic proximity to participating retailers may explain some of this difference in redemption rates: if participating retailers are not readily accessible, households may redeem fewer of the benefits issued. For reasons discussed in the body of this paper, the Summer EBT for Children (SEBTC) Demonstration provides a valuable opportunity to explore distance to retailers as an explanation for incomplete redemption and the SNAP/WIC differential in redemption rates. Using data from participants in the 2012 SEBTC sites, analyses reported in this paper do not support the hypothesis that retailer proximity is the cause of variation in redemption rates between participants who received benefits through the SNAP versus WIC programs. The paper also reports results of some analyses consistent with the conjecture that more limited selection of WIC-allowable foods may explain that program’s lower redemption rates. The balance of this paper proceeds as follows. The following section provides context for the analyses with a brief review of the literature on food deserts. The next section describes the SEBTC Demonstration, the available data (including descriptive statistics), and why SEBTC provides a valuable opportunity to explore the relationship between proximity to retailers and redemption rates. The subsequent section reports results of regression analyses of how redemptions vary with distance to food retailers. The results do not support the hypothesis that longer average distances travelled to redeem WIC benefits contributes to lower rates of redemption in WIC compared to SNAP. Rather, the results suggest that redemption rates are only very weakly related to a household’s distance to closest participating supermarket/superstore or to any participating retailer. The following section reports results for where redemptions actually occur and their implications for explaining SNAP/WIC differences. The next section presents analyses of redemption rates by WIC food categories, consistent with the hypothesis that lower WIC redemption rates are related in part to a lack of interest in certain WIC foods among program participants. The final section summarizes the results and discusses their implications for both policy and broader food desert literature. Appendix A describes database construction in more detail and reports the results of some sensitivity analyses with respect to database construction choices. Appendix B details our methods and provides additional substantive results. Policy Background The USDA Food and Nutrition Service (USDA/FNS) provides food assistance directly to households through SNAP and WIC. These two national programs with similar goals differ in important ways. The larger of the two programs, SNAP, provides low-income households with resources to purchase almost any foods (the major exceptions are hot prepared foods, alcohol, and tobacco) at numerous food retailers. SNAP benefits are issued through an Electronic Benefit Transfer (EBT) card, which operates like a standard consumer debit card. WIC is more narrowly targeted to low-income women who are pregnant or who care for children up to age five. The WIC program is moving towards the distribution of benefits using an EBT card; some states are distributing WIC via EBT, while others are not. Unlike SNAP participants, WIC participants must have certification from a health care professional that they are at nutritional risk. WIC benefits are accepted at fewer retailers than are SNAP benefits and can be spent only on specified quantities of approved foods. The WIC food package for children comprises nine food categories: juice, milk, breakfast cereal, cheese, eggs, fruits and vegetables, whole wheat bread, canned fish, and legumes or peanut butter. Given the restrictions on what can be purchased and at which retailers, it is not surprising that national participation rates among eligible persons are lower for WIC than for SNAP. Indeed, 83% of persons meeting SNAP eligibility guidelines participate in the program versus 63% of those meeting WIC requirements. Participation in the WIC program is even lower for eligible children aged 1 to 4 years old (54%; Johnson et al. 2015; Gray and Cunnyngham 2016). In addition, the limited available evidence suggests that redemption rates among participants are also lower in WIC as compared to SNAP; WIC participants redeem only 83% of their total benefits while SNAP participants redeem 97% of their benefits (Phillips et al. 2014 for three states; USDA 2011 nationally). At least four major differences between the SNAP and WIC programs exist, which may account for the differences in redemption rates among WIC participants. First, the WIC eligible population is demographically narrower and a broader range of income are eligible. WIC is available only to households with an infant, child, or pregnant or post-partum woman, with incomes below 185% of the federal poverty line (and that meet other eligibility requirements; USDA FNS 2017a). SNAP is available to all households with income below 130% of the federal poverty line (and that meet other eligibility requirements; see USDA FNS 2017b). It seems plausible that households with lower incomes have a greater need and would therefore be more likely to redeem more benefits. It is possible that the WIC package is better-suited to some demographic groups, such that redemption rates would be higher or lower. Second, the WIC benefit is smaller. In 2015, the average monthly SNAP benefit per person was $126.81 (USDA FNS 2016a). Valuing the foods at cost, average monthly cost per enrolled person in WIC was estimated as $43.37 in 2015 (USDA FNS 2016b), approximately one-third of the average monthly SNAP benefit. It seems plausible that because the benefit is smaller, beneficiaries would be more likely to decide that it was not worth the effort to redeem the benefit (any, more, and all). Third, WIC can only be spent on its narrow set of foods. As noted, choice of foods is relatively unrestricted in SNAP; choice of foods is strictly limited in WIC, to only the approved foods in the food package. Some households may prefer not to consume some of these WIC foods. Household members may also experience confusion regarding which foods qualify. Qualifying foods may not be available. Each of these factors would lead to lower redemption rates. Fourth, WIC operates through a narrower retailer network. Fewer retailers accept WIC than accept SNAP. Perhaps transportation issues imply that the lower density of WIC retailers leads to lower redemption rates. Differences in retailer networks are the focus of this article, which also provides some evidence on the impact of the narrow choice of foods on redemption. The hypothesis that the narrower network of WIC retailers may explain differences between SNAP and WIC participation and redemption is related to the broader literature on “food deserts,” that is, areas with limited access to supermarkets. The food desert literature explores the relationship between food retailer access and a range of dietary outcomes (Larson, Story, and Nelson 2009). Simple correlations suggest that residing in a food desert is related to poorer nutritional outcomes (see the reviews in Larson, Story, and Nelson 2009, and the Institute of Medicine and National Research Council 2009; see also USDA 2009). Crucially, the idea of food deserts implicitly assumes that people purchase the majority of their food at local retailers. That implicit assumption appears to be incorrect. For low-income families who participate in food assistance programs, available evidence suggests that households travel moderate distances to shop, often passing closer retailers—and even passing closer supermarkets/superstores—to shop at more-distant retailers (Edin et al. 2013; Mabli and Worthington 2015; ver Ploeg et al. 2015; Grindal et al. 2016; Schwartz et al. 2017). The evidence suggests that low-income households redeem their nutrition benefits at retailers considerably farther away from their residence rather than the closest participating retailer or participating supermarket/superstore. This evidence makes it less plausible that residing in a food desert (i.e., the lack of a supermarket near a household’s residence) would affect nutritional outcomes. Consistent with this interpretation, both (a) more-formal causal analyses of the relation of proximity to retailers and nutritional intake and (b) more-detailed analyses of actual food shopping behavior suggest that the correlation of food deserts with poorer nutritional outcomes is not causal. That is, residing in a food desert does not seem to cause worse nutritional outcomes, and placing supermarkets in food deserts is unlikely to substantially improve nutrition (Kyureghian, Nayga, and Bhattacharya 2012; Cummins, Flint, and Matthews 2014; Alcott, Diamond, and Dube 2015; Dubowitz et al. 2015; Handbury, Rahkosky, and Schnell 2016; see also the review of the more recent literature in ver Ploeg and Wilde, forthcoming). Summer EBT for Children Demonstration Data for this study are drawn from the SEBTC evaluation. The SEBTC Demonstration was designed to improve children’s food security and nutritional status in the summer, when school-based meal programs operate on a much smaller scale. To address this lack of food assistance in selected demonstration sites, SEBTC provides nutritional assistance to families with children eligible for free and reduced-price meals during the school year. In conjunction with the demonstration, FNS funded a large random assignment evaluation. (See Collins et al. [2013] for details regarding the SEBTC Demonstration and results of the evaluation including a description of implementation and estimated impacts.) The demonstration and evaluation were implemented in the summers of 2011 through 2014 by 10 grantees in 16 communities. Because the number of participating sites and individuals vary over the course of the study and the study design changes slightly, this paper focuses on the 2012 sample only. The 2012 intervention involved the largest sample and the greatest number of sites (see Collins et al. [2013] for more on site selection and the sites themselves).1 Specifically, the program operated as follows. Potential sites applied to USDA/FNS. Selected sites were given the option of implementing SEBTC according to SNAP rules (SEBTC-SNAP) or according to WIC rules (SEBTC-WIC). In sites choosing SEBTC-SNAP, benefits were distributed through the SNAP EBT system and could be used to purchase almost any foods at SNAP retailers. Also, the benefit was set at $60 per participating child in the household. In sites choosing SEBTC-WIC, benefits were distributed through the WIC EBT system (only sites that had implemented WIC EBT could choose this option). The SEBTC-WIC package was specified to include foods appropriate for school-age children, with a value of approximately $60 per participating child in the household. Benefits could only be redeemed at WIC retailers (many SNAP retailers do not participate in WIC, but almost all WIC retailers participate in SNAP). Thus, SEBTC-SNAP versus SEBTC-WIC was not randomly assigned. However, because store locations and stores’ decisions to accept SNAP and/or WIC were determined independently of a site’s decision to administer SNAP or WIC benefits, we do not believe this influences our analysis. Households provided consent (active consent in some sites, passive consent in other sites) to participate in the study. Lists of children in consenting households were forwarded to the evaluation, which randomly assigned all children in a household to either treatment or control (i.e., in a single household either all eligible children were assigned to treatment or all eligible children were assigned to control). This paper analyzes households assigned to the treatment condition and a benefit valued at about $60 per eligible child. We do not use the data or outcomes of households who received no benefit. This contrast makes SEBTC a powerful context in which to understand differences in redemption rates of the SNAP and WIC programs. In the previous section, we noted four factors that vary between the SNAP and WIC programs: (a) different populations, (b) different benefit amounts, (c) more-restricted choice of foods, and (d) proximity of retailer. Simple SNAP-WIC comparisons mix the impact of all four considerations. In contrast, analyses of SEBTC essentially eliminate the first and second sources of variation (i.e., different populations and different benefit amounts). Across SEBTC-SNAP and SEBTC-WIC sites, the individual eligibility conditions are the same (children eligible for free and reduced-price school meals) as are the benefit amounts per eligible child.2 Thus, these comparisons of SEBTC-SNAP and SEBTC-WIC isolate the effect of the third and fourth sources of variation (i.e., more restricted choices of food and retailer proximity). This paper examines how redemption adjusts with variation in retailer proximity (the fourth reason for SNAP/WIC difference in redemption rates), measured by the distance to the closest supermarket or superstore where participants could redeem benefits. Any remaining differences are due to differences between the specific grantees or sites, restricted choice (the third explanation), or chance. Specifically, we construct our analysis file as follows. We begin with data on every household assigned to the treatment group and all SNAP and WIC retailers in each of the nine states that participated in SEBTC in 2013. We then geocode all households and all food retailers using ArcGIS Online Geocoding Service and compute the road network driving distance to the closest retailer—WIC, SNAP, any retailer, and supermarket/superstore, a total of four distances. As discussed in appendix A, we exclude households from the analyses for the following reasons: insufficient physical address data, excessive distance to closest retailer (more than 10 miles), and location within one mile of state border. Our data have few such households (733, or 2.1%). We also dropped two tribal sites (Cherokee and Chickasaw Nations) from the main analysis because a high proportion of households in these sites did not have physical addresses. In particular, the 10-mile cutoff was chosen to exclude only extreme observations. Few 2012 SEBTC households resided more than 10 miles from a participating retailer; from a total sample of about 30,000, the number of households that resided more than 10 miles from a participating retailer was only 115, and 73 of these households were in tribal sites. We analyzed the sensitivity of the paper’s results to these sample selection rules (see appendix B for details). Broadly speaking, the results are not sensitive to the inclusion of Cherokee and Chickasaw nation households or households that were excluded because of excessive distance to the nearest retailer or close proximity to the state border. In addition, we report site-specific results for the tribal sites. For analyses of redemption by retailer where SEBTC benefits were redeemed, we built a file with dollars redeemed by household by specific retailer. We then geocoded the distances from the household’s address to each of those retailers (see appendix A for details). Finally, for the restricted-choice analyses, we appended the dollar value of benefits redeemed in each WIC food category and analyzed the relationship to the nearest supermarket and redemption rates by food category. Table 2 presents sample statistics for the dependent variables. Considering the entire period in which the benefit was available in each site, redemption is nearly complete among SEBTC-SNAP households (94%) and substantially lower among SEBTC-WIC households (58%). Among those who redeemed any of their benefits (96% of SNAP households and 84% of WIC households), the average redemption rate was 88% overall, 98% for SNAP sites, and 69% for WIC sites. This finding of higher redemption rates for SEBTC-SNAP than for SEBTC-WIC is consistent with the finding of higher redemption rates for (conventional) SNAP than for (conventional) WIC. Table 1 Sample Statistics of Dependent Variables N Mean SD SNAP sites, 8 sites Redeemed any benefit 16,577 96.1% 19.5% Percent of benefit redeemed 16,577 93.9% 20.6% Percent of benefit redeemed (conditional on redeeming any benefit) 15,923 97.8% 8.0% Exhausted benefits in at least one month 16,577 81.1% 39.2% Exhausted benefits in all months 16,577 28.7% 45.2% WIC sites, 4 sites Redeemed any benefit 9,993 83.6% 37.0% Percent of benefit redeemed 9,993 57.8% 33.4% Percent of benefit redeemed (conditional on redeeming any benefit) 8,359 69.1% 23.5% Exhausted benefits in at least one month 9,993 13.3% 34.0% Exhausted benefits in all months 9,993 1.1% 10.3% N Mean SD SNAP sites, 8 sites Redeemed any benefit 16,577 96.1% 19.5% Percent of benefit redeemed 16,577 93.9% 20.6% Percent of benefit redeemed (conditional on redeeming any benefit) 15,923 97.8% 8.0% Exhausted benefits in at least one month 16,577 81.1% 39.2% Exhausted benefits in all months 16,577 28.7% 45.2% WIC sites, 4 sites Redeemed any benefit 9,993 83.6% 37.0% Percent of benefit redeemed 9,993 57.8% 33.4% Percent of benefit redeemed (conditional on redeeming any benefit) 8,359 69.1% 23.5% Exhausted benefits in at least one month 9,993 13.3% 34.0% Exhausted benefits in all months 9,993 1.1% 10.3% Source: SEBTC data. Table 1 Sample Statistics of Dependent Variables N Mean SD SNAP sites, 8 sites Redeemed any benefit 16,577 96.1% 19.5% Percent of benefit redeemed 16,577 93.9% 20.6% Percent of benefit redeemed (conditional on redeeming any benefit) 15,923 97.8% 8.0% Exhausted benefits in at least one month 16,577 81.1% 39.2% Exhausted benefits in all months 16,577 28.7% 45.2% WIC sites, 4 sites Redeemed any benefit 9,993 83.6% 37.0% Percent of benefit redeemed 9,993 57.8% 33.4% Percent of benefit redeemed (conditional on redeeming any benefit) 8,359 69.1% 23.5% Exhausted benefits in at least one month 9,993 13.3% 34.0% Exhausted benefits in all months 9,993 1.1% 10.3% N Mean SD SNAP sites, 8 sites Redeemed any benefit 16,577 96.1% 19.5% Percent of benefit redeemed 16,577 93.9% 20.6% Percent of benefit redeemed (conditional on redeeming any benefit) 15,923 97.8% 8.0% Exhausted benefits in at least one month 16,577 81.1% 39.2% Exhausted benefits in all months 16,577 28.7% 45.2% WIC sites, 4 sites Redeemed any benefit 9,993 83.6% 37.0% Percent of benefit redeemed 9,993 57.8% 33.4% Percent of benefit redeemed (conditional on redeeming any benefit) 8,359 69.1% 23.5% Exhausted benefits in at least one month 9,993 13.3% 34.0% Exhausted benefits in all months 9,993 1.1% 10.3% Source: SEBTC data. Table 2 Distance to Retailers Median Mean SD Min. 25th Percentile 75th Percentile Max. SNAP sites, 8 sites Driving distance to the nearest SNAP-accepting supermarket 1.00 1.28 1.14 0.00 0.66 1.47 9.96 Driving distance to the nearest SNAP-accepting retailer of any type 0.37 0.55 0.63 0.00 0.21 0.64 9.76 Driving distance to the nearest WIC-accepting supermarket 1.14 1.45 1.29 0.00 0.76 1.70 17.36 Driving distance to the nearest WIC-accepting retailer of any type 1.01 1.29 1.20 0.00 0.62 1.56 17.36 WIC sites, 4 sites Driving distance to the nearest SNAP-accepting supermarket 0.99 1.48 1.62 0.01 0.61 1.59 10.00 Driving distance to the nearest SNAP-accepting retailer of any type 0.38 0.63 0.87 0.00 0.21 0.67 9.06 Driving distance to the nearest WIC-accepting supermarket 1.05 1.56 1.65 0.01 0.64 1.70 10.00 Driving distance to the nearest WIC-accepting retailer of any type 0.91 1.36 1.53 0.00 0.46 1.59 9.98 Median Mean SD Min. 25th Percentile 75th Percentile Max. SNAP sites, 8 sites Driving distance to the nearest SNAP-accepting supermarket 1.00 1.28 1.14 0.00 0.66 1.47 9.96 Driving distance to the nearest SNAP-accepting retailer of any type 0.37 0.55 0.63 0.00 0.21 0.64 9.76 Driving distance to the nearest WIC-accepting supermarket 1.14 1.45 1.29 0.00 0.76 1.70 17.36 Driving distance to the nearest WIC-accepting retailer of any type 1.01 1.29 1.20 0.00 0.62 1.56 17.36 WIC sites, 4 sites Driving distance to the nearest SNAP-accepting supermarket 0.99 1.48 1.62 0.01 0.61 1.59 10.00 Driving distance to the nearest SNAP-accepting retailer of any type 0.38 0.63 0.87 0.00 0.21 0.67 9.06 Driving distance to the nearest WIC-accepting supermarket 1.05 1.56 1.65 0.01 0.64 1.70 10.00 Driving distance to the nearest WIC-accepting retailer of any type 0.91 1.36 1.53 0.00 0.46 1.59 9.98 Source: SEBTC data, FNS. Table 2 Distance to Retailers Median Mean SD Min. 25th Percentile 75th Percentile Max. SNAP sites, 8 sites Driving distance to the nearest SNAP-accepting supermarket 1.00 1.28 1.14 0.00 0.66 1.47 9.96 Driving distance to the nearest SNAP-accepting retailer of any type 0.37 0.55 0.63 0.00 0.21 0.64 9.76 Driving distance to the nearest WIC-accepting supermarket 1.14 1.45 1.29 0.00 0.76 1.70 17.36 Driving distance to the nearest WIC-accepting retailer of any type 1.01 1.29 1.20 0.00 0.62 1.56 17.36 WIC sites, 4 sites Driving distance to the nearest SNAP-accepting supermarket 0.99 1.48 1.62 0.01 0.61 1.59 10.00 Driving distance to the nearest SNAP-accepting retailer of any type 0.38 0.63 0.87 0.00 0.21 0.67 9.06 Driving distance to the nearest WIC-accepting supermarket 1.05 1.56 1.65 0.01 0.64 1.70 10.00 Driving distance to the nearest WIC-accepting retailer of any type 0.91 1.36 1.53 0.00 0.46 1.59 9.98 Median Mean SD Min. 25th Percentile 75th Percentile Max. SNAP sites, 8 sites Driving distance to the nearest SNAP-accepting supermarket 1.00 1.28 1.14 0.00 0.66 1.47 9.96 Driving distance to the nearest SNAP-accepting retailer of any type 0.37 0.55 0.63 0.00 0.21 0.64 9.76 Driving distance to the nearest WIC-accepting supermarket 1.14 1.45 1.29 0.00 0.76 1.70 17.36 Driving distance to the nearest WIC-accepting retailer of any type 1.01 1.29 1.20 0.00 0.62 1.56 17.36 WIC sites, 4 sites Driving distance to the nearest SNAP-accepting supermarket 0.99 1.48 1.62 0.01 0.61 1.59 10.00 Driving distance to the nearest SNAP-accepting retailer of any type 0.38 0.63 0.87 0.00 0.21 0.67 9.06 Driving distance to the nearest WIC-accepting supermarket 1.05 1.56 1.65 0.01 0.64 1.70 10.00 Driving distance to the nearest WIC-accepting retailer of any type 0.91 1.36 1.53 0.00 0.46 1.59 9.98 Source: SEBTC data, FNS. Table 2 presents descriptive statistics for the key independent variable, distance to closest retailer in various classes. Distance (in miles) to closest participating retailer—SNAP retailers in SEBTC-SNAP sites and WIC retailers in SEBTC-WIC sites—is shorter for SNAP sites (0.37 median, 0.55 mean) than for WIC sites (0.91 median, 1.36 mean). Differences are similar but less stark for distance to a participating supermarket or superstore (1.00 median, 1.28 mean for SEBTC-SNAP versus 1.05 median, 1.56 mean for SEBTC-WIC). Because distance is not randomly assigned and we are interested in plausibly causal interpretation of the analyses, we control for the following household characteristics: number of children in the household, age of the oldest child, race/ethnicity of the head of household, free or reduced lunch price eligibility, and household familial structure. We also include census tract-level regressors, including population density, median household income, population with a high school degree, population with no access to a vehicle, and population below the federal poverty line. We consider four specifications: (a) no controls (i.e., regressors), (b) site indicators, (c) site indicators, plus household characteristics, and (d) site indicators, characteristics of the sample household itself, and households’ characteristics for the census tract (computed from the five-year American Community Survey tabulations for 2012). Table 3 lists these controls and their sample statistics. Table 3 Sample Statistics of Household Covariates Household N Mean SD SNAP sites, 8 sites Household characteristics Number of children in household 16,577 1.8 1.0 Head of household is Black, non-Hispanic 11,654 46.0% 49.8% Head of household is Hispanic 11,654 11.7% 32.2% Household children are eligible for free lunch 13,235 91.9% 27.3% Household children are eligible for reduced-price lunch 13,235 9.7% 29.5% Oldest child is under 21 12,136 100.0% 1.6% Household is headed by a single female 13,728 33.7% 47.3% Household is headed by a single male 13,728 3.7% 19.0% Census tract characteristics Population density (persons per square mile) 16,571 4,524 3,956 Median income 16,571 $43,824 $18,493 Population with high school degree 16,571 82.0% 9.7% Population with no access to a vehicle 16,571 14.2% 12.2% Population below the federal poverty line 16,571 22.3% 13.8% WIC sites, 4 sites Household characteristics Number of children in household 9,993 1.8 1.0 Head of household is Black, non-Hispanic 7,893 10.4% 30.6% Head of household is Hispanic 7,893 48.7% 50.0% Household children are eligible for free lunch 5,027 91.0% 28.6% Household children are eligible for reduced-price lunch 5,027 11.5% 31.9% Oldest child is under 21 7,911 99.5% 7.1% Household is headed by a single female 7,913 20.3% 40.2% Household is headed by a single male 7,913 4.7% 21.2% Census tract characteristics Population density (persons per square mile) 9,993 4,555 3,132 Median income 9,989 $39,192 $14,181 Population with high school degree 9,993 76.5% 13.9% Population with no access to a vehicle 9,993 10.2% 7.8% Population below the federal poverty line 9,993 25.4% 13.7% Household N Mean SD SNAP sites, 8 sites Household characteristics Number of children in household 16,577 1.8 1.0 Head of household is Black, non-Hispanic 11,654 46.0% 49.8% Head of household is Hispanic 11,654 11.7% 32.2% Household children are eligible for free lunch 13,235 91.9% 27.3% Household children are eligible for reduced-price lunch 13,235 9.7% 29.5% Oldest child is under 21 12,136 100.0% 1.6% Household is headed by a single female 13,728 33.7% 47.3% Household is headed by a single male 13,728 3.7% 19.0% Census tract characteristics Population density (persons per square mile) 16,571 4,524 3,956 Median income 16,571 $43,824 $18,493 Population with high school degree 16,571 82.0% 9.7% Population with no access to a vehicle 16,571 14.2% 12.2% Population below the federal poverty line 16,571 22.3% 13.8% WIC sites, 4 sites Household characteristics Number of children in household 9,993 1.8 1.0 Head of household is Black, non-Hispanic 7,893 10.4% 30.6% Head of household is Hispanic 7,893 48.7% 50.0% Household children are eligible for free lunch 5,027 91.0% 28.6% Household children are eligible for reduced-price lunch 5,027 11.5% 31.9% Oldest child is under 21 7,911 99.5% 7.1% Household is headed by a single female 7,913 20.3% 40.2% Household is headed by a single male 7,913 4.7% 21.2% Census tract characteristics Population density (persons per square mile) 9,993 4,555 3,132 Median income 9,989 $39,192 $14,181 Population with high school degree 9,993 76.5% 13.9% Population with no access to a vehicle 9,993 10.2% 7.8% Population below the federal poverty line 9,993 25.4% 13.7% Source: SEBTC data and appended American Community Survey census tract characteristics data. Table 3 Sample Statistics of Household Covariates Household N Mean SD SNAP sites, 8 sites Household characteristics Number of children in household 16,577 1.8 1.0 Head of household is Black, non-Hispanic 11,654 46.0% 49.8% Head of household is Hispanic 11,654 11.7% 32.2% Household children are eligible for free lunch 13,235 91.9% 27.3% Household children are eligible for reduced-price lunch 13,235 9.7% 29.5% Oldest child is under 21 12,136 100.0% 1.6% Household is headed by a single female 13,728 33.7% 47.3% Household is headed by a single male 13,728 3.7% 19.0% Census tract characteristics Population density (persons per square mile) 16,571 4,524 3,956 Median income 16,571 $43,824 $18,493 Population with high school degree 16,571 82.0% 9.7% Population with no access to a vehicle 16,571 14.2% 12.2% Population below the federal poverty line 16,571 22.3% 13.8% WIC sites, 4 sites Household characteristics Number of children in household 9,993 1.8 1.0 Head of household is Black, non-Hispanic 7,893 10.4% 30.6% Head of household is Hispanic 7,893 48.7% 50.0% Household children are eligible for free lunch 5,027 91.0% 28.6% Household children are eligible for reduced-price lunch 5,027 11.5% 31.9% Oldest child is under 21 7,911 99.5% 7.1% Household is headed by a single female 7,913 20.3% 40.2% Household is headed by a single male 7,913 4.7% 21.2% Census tract characteristics Population density (persons per square mile) 9,993 4,555 3,132 Median income 9,989 $39,192 $14,181 Population with high school degree 9,993 76.5% 13.9% Population with no access to a vehicle 9,993 10.2% 7.8% Population below the federal poverty line 9,993 25.4% 13.7% Household N Mean SD SNAP sites, 8 sites Household characteristics Number of children in household 16,577 1.8 1.0 Head of household is Black, non-Hispanic 11,654 46.0% 49.8% Head of household is Hispanic 11,654 11.7% 32.2% Household children are eligible for free lunch 13,235 91.9% 27.3% Household children are eligible for reduced-price lunch 13,235 9.7% 29.5% Oldest child is under 21 12,136 100.0% 1.6% Household is headed by a single female 13,728 33.7% 47.3% Household is headed by a single male 13,728 3.7% 19.0% Census tract characteristics Population density (persons per square mile) 16,571 4,524 3,956 Median income 16,571 $43,824 $18,493 Population with high school degree 16,571 82.0% 9.7% Population with no access to a vehicle 16,571 14.2% 12.2% Population below the federal poverty line 16,571 22.3% 13.8% WIC sites, 4 sites Household characteristics Number of children in household 9,993 1.8 1.0 Head of household is Black, non-Hispanic 7,893 10.4% 30.6% Head of household is Hispanic 7,893 48.7% 50.0% Household children are eligible for free lunch 5,027 91.0% 28.6% Household children are eligible for reduced-price lunch 5,027 11.5% 31.9% Oldest child is under 21 7,911 99.5% 7.1% Household is headed by a single female 7,913 20.3% 40.2% Household is headed by a single male 7,913 4.7% 21.2% Census tract characteristics Population density (persons per square mile) 9,993 4,555 3,132 Median income 9,989 $39,192 $14,181 Population with high school degree 9,993 76.5% 13.9% Population with no access to a vehicle 9,993 10.2% 7.8% Population below the federal poverty line 9,993 25.4% 13.7% Source: SEBTC data and appended American Community Survey census tract characteristics data. Table 4 Effects of Distance to the Closest Supermarket on Proportion Redeemed All Households SNAP Households WIC Households Driving distance to the nearest supermarket where benefits could be redeemed 0.000 −0.001 0.001 (0.001) (0.001) (0.002) Constant 0.605*** 0.878*** 0.680*** (0.050) (0.066) (0.074) R-squared 0.345 0.026 0.075 N 26,570 16,577 9,993 All Households SNAP Households WIC Households Driving distance to the nearest supermarket where benefits could be redeemed 0.000 −0.001 0.001 (0.001) (0.001) (0.002) Constant 0.605*** 0.878*** 0.680*** (0.050) (0.066) (0.074) R-squared 0.345 0.026 0.075 N 26,570 16,577 9,993 Note: Two-sided hypothesis test; asterisks *** = statistically significant p<0.01; ** = statistically significant, p<0.05; and * = statistically significant, p<0.10. Table 4 Effects of Distance to the Closest Supermarket on Proportion Redeemed All Households SNAP Households WIC Households Driving distance to the nearest supermarket where benefits could be redeemed 0.000 −0.001 0.001 (0.001) (0.001) (0.002) Constant 0.605*** 0.878*** 0.680*** (0.050) (0.066) (0.074) R-squared 0.345 0.026 0.075 N 26,570 16,577 9,993 All Households SNAP Households WIC Households Driving distance to the nearest supermarket where benefits could be redeemed 0.000 −0.001 0.001 (0.001) (0.001) (0.002) Constant 0.605*** 0.878*** 0.680*** (0.050) (0.066) (0.074) R-squared 0.345 0.026 0.075 N 26,570 16,577 9,993 Note: Two-sided hypothesis test; asterisks *** = statistically significant p<0.01; ** = statistically significant, p<0.05; and * = statistically significant, p<0.10. Analyses discussed in the body of the paper use the full set of controls. Appendix B explores the sensitivity of the paper’s main results to the specific controls included in the regression models. Once the site indicators are included (i.e., the second specification), we observe only minimal sensitivity (see appendix table B.1.1). The body of the article therefore only reports results for the fourth specification. Redemption and Distance to Retailers This section uses regression analysis to explore the relationship between distance to closest retailer and redemption rate, that is, benefits redeemed as a percentage of benefits issued. Analyses in appendix B show results for similar models for redemption of both any and all benefits. General patterns are similar. The generic regression is of the form: Y = β0+ β1Distance + β2ControlVars + β3Site +u (1) where Y represents our benefit redemption outcomes; Distance represents distance to some category of food retailer where a household could have redeemed its SEBTC benefits;3 in the primary specification, this is the distance to the nearest SNAP-accepting supermarket (SNAP sites) and to the nearest WIC-accepting supermarket (WIC sites); ControlVars is a vector of characteristics from the SEBTC sampling file and census tract; Site represents a dummy variable for each demonstration site (=1 for the site in which the household was located, and 0 otherwise); β0, β1, β2, and β3, are parameters to be estimated; u is a regression residual. For those analyses in which the benefit redemption outcomes are binary (e.g., take-up of any benefit and benefit exhaustion) these analyses are linear probability models. We run all analyses separately for WIC and SNAP sites. Table 4 shows the results of the regression analyses for SNAP households and WIC households separately, where “distance” is miles. These results imply that there is virtually no relationship between distance to the nearest supermarket where benefits could be redeemed and the proportion of benefits redeemed over the course of the summer of 2012. For SNAP sites there is a small negative (but not statistically significant) relationship between distance and benefits redeemed. A SNAP household that is 10 miles away from its nearest SNAP-accepting supermarket (the largest distance in our analysis file) would, on average, redeem 1 percentage point less of its benefits. The opposite is true when including only WIC households: a WIC household 10 miles4 away from the closest supermarket that accepted WIC benefits would, on average, redeem 1 percentage point more of its total benefits. Furthermore, the results are precise: 95% confidence intervals for SNAP sites are approximately -3% to -1%, and for WIC sites approximately −1% to 3%. Results are not sensitive to covariates (table B.1.1) for distance to closest retailer (rather than closest supermarket or superstore table B.1.2) and when including both distance to closest supermarket/superstore and to any retailer (table B.1.3). The previous model assumes that the change in redemption is constant with each additional mile to the closest retailer. To relax this assumption, we estimated linear spline models, which allow the effect of the first mile (or fraction thereof) to differ from the effect of miles past the first mile (Grindal et al. [2016] find some evidence for such a difference.) We also consider models in which the break point is at three-quarters of a mile (rather than a mile). Estimates from these spline models imply that redemption decreases slightly (and the relationship is statistically significant) for the first mile between the household and the retailer (about 1 percentage point), but there is no relationship if instead the knot is at 0.75 miles (see appendix table B.1.4). Considering each of the 12 participating sites separately, in a linear model, the effects of distance to the nearest supermarket are near 0% in each site (table 5). In some sites the effect is slightly negative, and in others it is slightly positive. In Delaware (a SNAP site) and Grand Rapids (a WIC site) we observe significant, negative relationships between distance and redemption rates. In Delaware, a participant redeemed 0.7% less of his or her benefits for each additional mile between a household and the nearest supermarket where they could redeem benefits. In Grand Rapids, participants, on average, redeemed nearly 2% less of their benefit amount for each additional mile. Again, these site-specific results should be interpreted with caution given the numerous parameters estimated. Perhaps distance matters in some sites, but not others. Alternatively, the results may be due to chance, as we see no conclusive patterns or site characteristics that would allow us to explain variations in our findings. Distance to Retailers Where Benefits Are Redeemed In the summer of 2012, there were approximately 50,000 authorized SNAP retailers in the nine states where the SEBTC sites were located. Only 7,186 of these retailers also accepted WIC (15%). Despite the large number of retailers who accept SNAP benefits, nearly two-thirds of SEBTC-SNAP redemptions (65% of dollars redeemed) in July of 2012 (the month when all demonstration sites participated) were redeemed at stores that also accepted WIC. This high proportion of redemptions in SEBTC-SNAP sites at WIC-accepting retailers is probably related in part to the fact that the overwhelming share of benefits are spent at supermarkets and superstores (86% of SEBTC redemptions in SNAP sites and 89% of SEBTC redemptions in WIC sites), and most supermarkets and superstores accept both WIC and SNAP. Of all SNAP-accepting supermarkets and superstores, 69% of SEBTC redemptions occur at supermarkets and superstores that accept WIC. Few SEBTC benefits were redeemed at participants’ closest participating retailer or even the closest participating supermarket. In WIC sites, 29% of redemptions were made at a household’s closest participating retailer (i.e., where any redemption was made), and 25% of redemptions were made at a household’s closest participating supermarket (sometimes the closest participating retailer is a supermarket). In SNAP sites, only 4% of redemptions were made at the closest participating retailer, and only 2% at the closest participating supermarket. Furthermore, it is not just that beneficiaries do not redeem at the closest retailer. In SNAP sites, the majority of redemptions (53%) were made at stores farther away than the 25 closest participating retailers. In both SNAP and WIC sites, participants are traveling approximately four times farther than the closest supermarket or superstore where they could redeem benefits. Weighting by benefits redeemed, the average distance traveled was 3.3 miles for an SEBTC-SNAP household and 4.6 miles for an SEBTC-WIC household. Similarly, the distance to the retailer at which the most benefits were redeemed was 3.9 miles for SEBTC-SNAP and 4.7 miles for SEBTC-WIC. Each of these distances is several times the mean distance to the closest participating retailer (0.6 miles for SEBTC-SNAP and 1.4 miles for SEBTC-WIC) or the closest participating supermarket/superstore (1.3 miles for SEBTC-SNAP and 1.6 miles for SEBTC-WIC). Results on Redemption by Specific Foods WIC participants can redeem benefits only for approved foods within the specific categories and at quantities of their food prescriptions. One hypothesis for lower redemption rates in WIC sites is that participants may not want to redeem some (or all of some) foods, even if they are available with minimal travel. It is either not perceived as worth the effort of carrying them home or the inconvenience of separating their purchases, or there is so little interest in the allowed foods. Other hypotheses are possible. Some of the foods are more perishable (milk, cheese, eggs, fruits and vegetables); others are less perishable (cereal, beans, canned fish). Because of the details of allowable WIC foods within a category, some foods are more challenging to obtain in acceptable forms (cereal, bread, juice) and in quantities that allow spending out a category. Table 6 shows the average dollar amounts spent in each category for each of the four WIC sites included in the analysis. Prices of groceries vary from site to site, and allowable quantities vary across food categories, so it is more informative to look at the within-category redemption rates. Table 7 shows the redemption rates for each WIC food category, calculated as the total cost of units purchased/total cost of units allowed for all foods covered by the WIC benefit in a specific category. Considering all WIC sites together and each site separately, redemption rates were generally lowest in the beans, fish, and bread categories. However, excluding households that redeemed none of their SEBTC-WIC benefits, redemption rates still do not approach the rates of nearly complete redemptions in the SNAP sites. Table 5 Effects of Distance to the Closest Supermarket/Superstore where Benefits Could Be Redeemed on Percentage Redeemed by Site Coefficient (SE) SNAP Sites Eastern Connecticut 0.003 (0.002) Western Connecticut 0.016 (0.012) Delaware −0.007* (0.004) Missouri-Kansas City −0.007 (0.010) Missouri-St. Louis −0.012 (0.010) Oregon-Deschutes −0.001 (0.002) Oregon-Salem 0.002 (0.004) Washington −0.001 (0.006) WIC Sites Cherokee −0.000 (0.004) Chickasaw 0.004 (0.003) Michigan-Eastern 0.004 (0.002) Michigan-Grand Rapids −0.019* (0.010) Nevada −0.000 (0.004) Texas −0.006 (0.010) Coefficient (SE) SNAP Sites Eastern Connecticut 0.003 (0.002) Western Connecticut 0.016 (0.012) Delaware −0.007* (0.004) Missouri-Kansas City −0.007 (0.010) Missouri-St. Louis −0.012 (0.010) Oregon-Deschutes −0.001 (0.002) Oregon-Salem 0.002 (0.004) Washington −0.001 (0.006) WIC Sites Cherokee −0.000 (0.004) Chickasaw 0.004 (0.003) Michigan-Eastern 0.004 (0.002) Michigan-Grand Rapids −0.019* (0.010) Nevada −0.000 (0.004) Texas −0.006 (0.010) Note: Two-sided hypothesis test; asterisks *** = statistically significant, p<0.01; ** = statistically significant, p<0.05; * = statistically significant, p<0.10. Table 5 Effects of Distance to the Closest Supermarket/Superstore where Benefits Could Be Redeemed on Percentage Redeemed by Site Coefficient (SE) SNAP Sites Eastern Connecticut 0.003 (0.002) Western Connecticut 0.016 (0.012) Delaware −0.007* (0.004) Missouri-Kansas City −0.007 (0.010) Missouri-St. Louis −0.012 (0.010) Oregon-Deschutes −0.001 (0.002) Oregon-Salem 0.002 (0.004) Washington −0.001 (0.006) WIC Sites Cherokee −0.000 (0.004) Chickasaw 0.004 (0.003) Michigan-Eastern 0.004 (0.002) Michigan-Grand Rapids −0.019* (0.010) Nevada −0.000 (0.004) Texas −0.006 (0.010) Coefficient (SE) SNAP Sites Eastern Connecticut 0.003 (0.002) Western Connecticut 0.016 (0.012) Delaware −0.007* (0.004) Missouri-Kansas City −0.007 (0.010) Missouri-St. Louis −0.012 (0.010) Oregon-Deschutes −0.001 (0.002) Oregon-Salem 0.002 (0.004) Washington −0.001 (0.006) WIC Sites Cherokee −0.000 (0.004) Chickasaw 0.004 (0.003) Michigan-Eastern 0.004 (0.002) Michigan-Grand Rapids −0.019* (0.010) Nevada −0.000 (0.004) Texas −0.006 (0.010) Note: Two-sided hypothesis test; asterisks *** = statistically significant, p<0.01; ** = statistically significant, p<0.05; * = statistically significant, p<0.10. Table 6 Average Dollar Amounts Redeemed, in Each WIC Category, over the Entire Summer Site N Total Bread Cheese Beans Cereal Eggs Fish Fruits/veg Juice Milk Michigan - Eastern 1,390 $438.36 $32.06 $37.29 $42.54 $49.75 $11.16 $16.11 $106.13 $23.48 $51.53 Michigan – Grand Rapids 2,323 $327.64 $20.15 $24.53 $19.81 $36.27 $7.25 $10.84 $71.71 $16.33 $41.24 NV 2,383 $375.48 $30.35 $18.40 $15.41 $28.77 $6.67 $11.87 $61.87 $12.81 $36.82 TX 2,421 $274.94 $22.62 $18.66 $19.79 $28.65 $7.53 $13.21 $67.49 10.86 40.45 Site N Total Bread Cheese Beans Cereal Eggs Fish Fruits/veg Juice Milk Michigan - Eastern 1,390 $438.36 $32.06 $37.29 $42.54 $49.75 $11.16 $16.11 $106.13 $23.48 $51.53 Michigan – Grand Rapids 2,323 $327.64 $20.15 $24.53 $19.81 $36.27 $7.25 $10.84 $71.71 $16.33 $41.24 NV 2,383 $375.48 $30.35 $18.40 $15.41 $28.77 $6.67 $11.87 $61.87 $12.81 $36.82 TX 2,421 $274.94 $22.62 $18.66 $19.79 $28.65 $7.53 $13.21 $67.49 10.86 40.45 Table 6 Average Dollar Amounts Redeemed, in Each WIC Category, over the Entire Summer Site N Total Bread Cheese Beans Cereal Eggs Fish Fruits/veg Juice Milk Michigan - Eastern 1,390 $438.36 $32.06 $37.29 $42.54 $49.75 $11.16 $16.11 $106.13 $23.48 $51.53 Michigan – Grand Rapids 2,323 $327.64 $20.15 $24.53 $19.81 $36.27 $7.25 $10.84 $71.71 $16.33 $41.24 NV 2,383 $375.48 $30.35 $18.40 $15.41 $28.77 $6.67 $11.87 $61.87 $12.81 $36.82 TX 2,421 $274.94 $22.62 $18.66 $19.79 $28.65 $7.53 $13.21 $67.49 10.86 40.45 Site N Total Bread Cheese Beans Cereal Eggs Fish Fruits/veg Juice Milk Michigan - Eastern 1,390 $438.36 $32.06 $37.29 $42.54 $49.75 $11.16 $16.11 $106.13 $23.48 $51.53 Michigan – Grand Rapids 2,323 $327.64 $20.15 $24.53 $19.81 $36.27 $7.25 $10.84 $71.71 $16.33 $41.24 NV 2,383 $375.48 $30.35 $18.40 $15.41 $28.77 $6.67 $11.87 $61.87 $12.81 $36.82 TX 2,421 $274.94 $22.62 $18.66 $19.79 $28.65 $7.53 $13.21 $67.49 10.86 40.45 Table 7 Average Redemption Rates by Food Category All WIC Sites Eastern Michigan Grand Rapids, Michigan Nevada Texas Unconditional Conditional Unconditional Conditional Unconditional Conditional Unconditional Conditional Unconditional Conditional N 9,988 8,359 1,422 1,350 2,773 2,532 2,887 2,346 2,906 2,131 Milk 65.8% 78.7% 75.8% 79.9% 70.3% 76.9% 57.8% 71.1% 64.7% 88.2% Cheese 66.3% 79.3% 76.6% 80.7% 68.7% 75.3% 58.0% 71.3% 67.4% 91.9% Eggs 66.1% 79.0% 74.2% 78.2% 75.1% 82.3% 51.5% 63.3% 68.1% 92.9% Juice 66.0% 78.9% 74.4% 78.4% 73.7% 80.7% 53.8% 66.2% 66.7% 90.9% Cereal 60.6% 72.4% 64.3% 67.8% 62.4% 68.3% 52.0% 64.0% 65.6% 89.5% Beans 53.9% 64.3% 67.9% 71.5% 55.7% 61.0% 37.4% 46.0% 61.6% 84.0% Fish 52.1% 62.2% 70.6% 74.3% 57.2% 62.7% 33.2% 40.9% 56.9% 77.6% Bread 48.1% 57.5% 42.0% 44.2% 41.9% 45.9% 43.0% 52.9% 62.2% 84.8% Fruits/ veg 64.4% 76.9% 68.1% 71.8% 72.3% 79.2% 54.7% 67.3% 64.5% 88.0% All WIC Sites Eastern Michigan Grand Rapids, Michigan Nevada Texas Unconditional Conditional Unconditional Conditional Unconditional Conditional Unconditional Conditional Unconditional Conditional N 9,988 8,359 1,422 1,350 2,773 2,532 2,887 2,346 2,906 2,131 Milk 65.8% 78.7% 75.8% 79.9% 70.3% 76.9% 57.8% 71.1% 64.7% 88.2% Cheese 66.3% 79.3% 76.6% 80.7% 68.7% 75.3% 58.0% 71.3% 67.4% 91.9% Eggs 66.1% 79.0% 74.2% 78.2% 75.1% 82.3% 51.5% 63.3% 68.1% 92.9% Juice 66.0% 78.9% 74.4% 78.4% 73.7% 80.7% 53.8% 66.2% 66.7% 90.9% Cereal 60.6% 72.4% 64.3% 67.8% 62.4% 68.3% 52.0% 64.0% 65.6% 89.5% Beans 53.9% 64.3% 67.9% 71.5% 55.7% 61.0% 37.4% 46.0% 61.6% 84.0% Fish 52.1% 62.2% 70.6% 74.3% 57.2% 62.7% 33.2% 40.9% 56.9% 77.6% Bread 48.1% 57.5% 42.0% 44.2% 41.9% 45.9% 43.0% 52.9% 62.2% 84.8% Fruits/ veg 64.4% 76.9% 68.1% 71.8% 72.3% 79.2% 54.7% 67.3% 64.5% 88.0% Note: “Unconditional”—among everyone, whether or not any benefits were redeemed; “Conditional”—among those who redeemed any benefits. Table 7 Average Redemption Rates by Food Category All WIC Sites Eastern Michigan Grand Rapids, Michigan Nevada Texas Unconditional Conditional Unconditional Conditional Unconditional Conditional Unconditional Conditional Unconditional Conditional N 9,988 8,359 1,422 1,350 2,773 2,532 2,887 2,346 2,906 2,131 Milk 65.8% 78.7% 75.8% 79.9% 70.3% 76.9% 57.8% 71.1% 64.7% 88.2% Cheese 66.3% 79.3% 76.6% 80.7% 68.7% 75.3% 58.0% 71.3% 67.4% 91.9% Eggs 66.1% 79.0% 74.2% 78.2% 75.1% 82.3% 51.5% 63.3% 68.1% 92.9% Juice 66.0% 78.9% 74.4% 78.4% 73.7% 80.7% 53.8% 66.2% 66.7% 90.9% Cereal 60.6% 72.4% 64.3% 67.8% 62.4% 68.3% 52.0% 64.0% 65.6% 89.5% Beans 53.9% 64.3% 67.9% 71.5% 55.7% 61.0% 37.4% 46.0% 61.6% 84.0% Fish 52.1% 62.2% 70.6% 74.3% 57.2% 62.7% 33.2% 40.9% 56.9% 77.6% Bread 48.1% 57.5% 42.0% 44.2% 41.9% 45.9% 43.0% 52.9% 62.2% 84.8% Fruits/ veg 64.4% 76.9% 68.1% 71.8% 72.3% 79.2% 54.7% 67.3% 64.5% 88.0% All WIC Sites Eastern Michigan Grand Rapids, Michigan Nevada Texas Unconditional Conditional Unconditional Conditional Unconditional Conditional Unconditional Conditional Unconditional Conditional N 9,988 8,359 1,422 1,350 2,773 2,532 2,887 2,346 2,906 2,131 Milk 65.8% 78.7% 75.8% 79.9% 70.3% 76.9% 57.8% 71.1% 64.7% 88.2% Cheese 66.3% 79.3% 76.6% 80.7% 68.7% 75.3% 58.0% 71.3% 67.4% 91.9% Eggs 66.1% 79.0% 74.2% 78.2% 75.1% 82.3% 51.5% 63.3% 68.1% 92.9% Juice 66.0% 78.9% 74.4% 78.4% 73.7% 80.7% 53.8% 66.2% 66.7% 90.9% Cereal 60.6% 72.4% 64.3% 67.8% 62.4% 68.3% 52.0% 64.0% 65.6% 89.5% Beans 53.9% 64.3% 67.9% 71.5% 55.7% 61.0% 37.4% 46.0% 61.6% 84.0% Fish 52.1% 62.2% 70.6% 74.3% 57.2% 62.7% 33.2% 40.9% 56.9% 77.6% Bread 48.1% 57.5% 42.0% 44.2% 41.9% 45.9% 43.0% 52.9% 62.2% 84.8% Fruits/ veg 64.4% 76.9% 68.1% 71.8% 72.3% 79.2% 54.7% 67.3% 64.5% 88.0% Note: “Unconditional”—among everyone, whether or not any benefits were redeemed; “Conditional”—among those who redeemed any benefits. Table 8 Effects of Distance to the Closest Supermarket on percentage Redeemed by WIC Category All WIC Sites Eastern Michigan Grand Rapids, Michigan Nevada Texas Unconditional Conditional Unconditional Conditional Unconditional Conditional Unconditional Conditional Unconditional Conditional Milk 0.002 0.001 0.007** 0.004 −0.029** −0.027** 0 0.001 −0.006 −0.005 Cheese −0.001 −0.001 0.004 0.001 −0.01 −0.006 −0.001 −0.001 −0.007 −0.006 Eggs −0.004* −0.006*** 0 −0.002 −0.013 −0.009 −0.005 −0.006** −0.004 −0.002 Juice 0.002 0.002 0.007* 0.004 −0.033** −0.030*** 0.001 0.002 −0.004 −0.002 Cereal 0.002 0.002 0.004 0.002 −0.033*** −0.031*** 0.002 0.003 −0.004 −0.001 Beans 0 −0.001 0.002 0 −0.003 0.001 −0.001 −0.001 −0.011 −0.011 Fish 0.002 0.001 0.005 0.002 −0.008 −0.004 0 0 −0.009 −0.008 Bread 0.003 0.003 0.007** 0.006* −0.006 −0.003 −0.001 0 −0.003 −0.001 Fruits/veg 0 −0.001 0 −0.002 −0.014 −0.011 0 0.001 −0.003 −0.001 All WIC Sites Eastern Michigan Grand Rapids, Michigan Nevada Texas Unconditional Conditional Unconditional Conditional Unconditional Conditional Unconditional Conditional Unconditional Conditional Milk 0.002 0.001 0.007** 0.004 −0.029** −0.027** 0 0.001 −0.006 −0.005 Cheese −0.001 −0.001 0.004 0.001 −0.01 −0.006 −0.001 −0.001 −0.007 −0.006 Eggs −0.004* −0.006*** 0 −0.002 −0.013 −0.009 −0.005 −0.006** −0.004 −0.002 Juice 0.002 0.002 0.007* 0.004 −0.033** −0.030*** 0.001 0.002 −0.004 −0.002 Cereal 0.002 0.002 0.004 0.002 −0.033*** −0.031*** 0.002 0.003 −0.004 −0.001 Beans 0 −0.001 0.002 0 −0.003 0.001 −0.001 −0.001 −0.011 −0.011 Fish 0.002 0.001 0.005 0.002 −0.008 −0.004 0 0 −0.009 −0.008 Bread 0.003 0.003 0.007** 0.006* −0.006 −0.003 −0.001 0 −0.003 −0.001 Fruits/veg 0 −0.001 0 −0.002 −0.014 −0.011 0 0.001 −0.003 −0.001 Note: Two-sided hypothesis test; asterisks *** = statistically significant, p<0.01; ** = statistically significant, p<0.05; * = statistically significant, p<0.10. Table 8 Effects of Distance to the Closest Supermarket on percentage Redeemed by WIC Category All WIC Sites Eastern Michigan Grand Rapids, Michigan Nevada Texas Unconditional Conditional Unconditional Conditional Unconditional Conditional Unconditional Conditional Unconditional Conditional Milk 0.002 0.001 0.007** 0.004 −0.029** −0.027** 0 0.001 −0.006 −0.005 Cheese −0.001 −0.001 0.004 0.001 −0.01 −0.006 −0.001 −0.001 −0.007 −0.006 Eggs −0.004* −0.006*** 0 −0.002 −0.013 −0.009 −0.005 −0.006** −0.004 −0.002 Juice 0.002 0.002 0.007* 0.004 −0.033** −0.030*** 0.001 0.002 −0.004 −0.002 Cereal 0.002 0.002 0.004 0.002 −0.033*** −0.031*** 0.002 0.003 −0.004 −0.001 Beans 0 −0.001 0.002 0 −0.003 0.001 −0.001 −0.001 −0.011 −0.011 Fish 0.002 0.001 0.005 0.002 −0.008 −0.004 0 0 −0.009 −0.008 Bread 0.003 0.003 0.007** 0.006* −0.006 −0.003 −0.001 0 −0.003 −0.001 Fruits/veg 0 −0.001 0 −0.002 −0.014 −0.011 0 0.001 −0.003 −0.001 All WIC Sites Eastern Michigan Grand Rapids, Michigan Nevada Texas Unconditional Conditional Unconditional Conditional Unconditional Conditional Unconditional Conditional Unconditional Conditional Milk 0.002 0.001 0.007** 0.004 −0.029** −0.027** 0 0.001 −0.006 −0.005 Cheese −0.001 −0.001 0.004 0.001 −0.01 −0.006 −0.001 −0.001 −0.007 −0.006 Eggs −0.004* −0.006*** 0 −0.002 −0.013 −0.009 −0.005 −0.006** −0.004 −0.002 Juice 0.002 0.002 0.007* 0.004 −0.033** −0.030*** 0.001 0.002 −0.004 −0.002 Cereal 0.002 0.002 0.004 0.002 −0.033*** −0.031*** 0.002 0.003 −0.004 −0.001 Beans 0 −0.001 0.002 0 −0.003 0.001 −0.001 −0.001 −0.011 −0.011 Fish 0.002 0.001 0.005 0.002 −0.008 −0.004 0 0 −0.009 −0.008 Bread 0.003 0.003 0.007** 0.006* −0.006 −0.003 −0.001 0 −0.003 −0.001 Fruits/veg 0 −0.001 0 −0.002 −0.014 −0.011 0 0.001 −0.003 −0.001 Note: Two-sided hypothesis test; asterisks *** = statistically significant, p<0.01; ** = statistically significant, p<0.05; * = statistically significant, p<0.10. We also explored whether some food categories are more sensitive to distance (e.g., perishables or high weight to value). Table 8 shows the change in proportion of benefits redeemed in each category for each additional mile between the household and the supermarket where they could redeem benefits, both modeling unconditional redemption (i.e., including those who redeem no benefits) and conditional redemption (i.e., excluding those who redeem no benefits). Here, as in the rest of the paper, we focus on the unconditional analysis. Again, considering all WIC sites together and each site separately, the farther the distance to the nearest supermarket, the lower the redemption rate for eggs benefits. In Grand Rapids, households that are farther away from their nearest supermarket or superstore redeem statistically significantly less milk, juice, and cereal (but not eggs). In Eastern Michigan, households that are farther away redeem statistically significantly more milk, bread, and juice. The lack of any relation in two of the sites and the unexpected sign in one of the sites lead us to infer that the observed relation is spurious, probably due to multiple comparisons issues (i.e., we are testing a lot of hypotheses, some of them will appear significant simply by chance). As with the site-specific results, these food category-specific results should be interpreted with caution. Perhaps distance matters for some food categories, but not for others. Alternatively, the results may be due to chance. Discussion Among U.S. nutrition assistance programs, redemption rates are much higher for SNAP benefits than for WIC benefits. Some prior work suggests that living farther away from supermarkets may create barriers to a family’s ability to access affordable healthy food. Work by Rose and Richards (2004) found that a greater distance to supermarkets was associated with significantly lower fruit consumption among SNAP participants. Other work suggests that opening a new supermarket in a food desert can lead to increased fruit and vegetable consumption for low-income families (Cummins, Flint, and Matthews 2014; Elbel et al. 2017). This paper provides three results on SEBTC redemption and distance to retailers, and more broadly on distance to retailers as an explanation of why redemption rates are lower for WIC than for SNAP. First, the relationship between distance and redemption rates is quite small. Our results imply that even if every SNAP retailer also accepted WIC, redemption rates for WIC-participating families would not meaningfully increase. Similarly, if every SEBTC-WIC household had a participating retailer next door, redemption rates would barely change. Second, our analysis of the distances households travel to redeem benefits does not support the hypothesis that the longer average distance to WIC retailers represents an important explanation of lower WIC redemption rates. Weighting by dollars redeemed, on average SEBTC-SNAP households redeem their benefits at retailers 6.3 times as far away as the closest participating retailer, and 2.7 times as far as the closest supermarket/superstore. For SEBTC-WIC households, the corresponding figures are slightly smaller: 3.0 times and 2.5 times, respectively. Third, disaggregating WIC redemption rates by broad food category suggests that, with the exception of eggs, for no food category is redemption strongly related to distance. In contrast, redemption varies strongly with food category. In-depth qualitative interviews with households and a survey might provide insight into the relation between WIC’s restrictive food choices and lower redemption rates. It is possible that participants are confused by the various restrictions, or the WIC packages may not be appropriately constructed to meet participant needs. However, our results are inconsistent with any role of proximity to a retailer, even for perishable items. Together these results suggest that distance to retailers was not a major consideration in benefit redemption for Summer EBT households in 2012. Given that these results suggest that households do not shop at the closest retailer or even supermarket/superstore, it seems unlikely that distance to closest retailer or supermarket/superstore represents an important causal factor for the overwhelming share of households. These results are broadly consistent with recent developments in the “food desert” literature suggesting that food desert correlations—that is, that nutritional outcomes are worse in food deserts—are not causal. However, our sample does not contain participants in more rural areas with longer distances to retailers, so our results are not generalizable to this population. Our analyses also only include households with school-age children. It is possible that redemption patterns and the sensitivity to distance is different for households without school-age children (e.g., households that only have younger children, households with no children, single-person households). Footnotes 1 We chose to focus on one year of data to account for the fact that retailer options and location change over time. In this case, a cross sectional study is most appropriate. The study year 2012 provided us the largest sample and the simplest study design. For example, in 2013, households received $60 or $30 in benefits, but no households received zero benefits. 2 While whether a site adopted SEBTC-SNAP or SEBTC-WIC was not randomly assigned, our analyses control for observed differences in households; see below. Nevertheless, the small number of sites means that the results should be interpreted with caution. 3 Retailers include all stores where participants can redeem benefits. In addition to supermarkets and superstores, this may also include specialty stores such as butchers or bakeries, wholesalers, and convenience stores. Because the majority of redemptions happen at supermarkets and superstores, we limit our main analyses to these categories. 4 Ten miles is the maximum distance between a household and a retailer. See appendix A for sample deletion decisions. References Allcott H. , Diamond R. , Dube J.P . 2015 . The Geography of Poverty and Nutrition: Food Deserts and Food Choices across the U.S. New York University Working Paper. Castner L. , Henke J . 2011 . Benefit Redemption Patterns in the Supplemental Nutrition Assistance Program (No. b746c9a56cb34547b475799386b0182a). Washington, D.C: Mathematica Policy Research. Collins A. , Briefel R. , Klerman J.A. , Rowe G. , Wolf A. , Logan C.W. , Gordon A. et al. 2013 . Summer Electronic Benefits Transfer for Children (SEBTC) Demonstration: 2012 Final Report. Nutrition Assistance Program Report. Washington DC: U.S. Department of Agriculture, Food and Nutrition Service, Office of Policy Support. Cummins S. , Flint E. , Matthews S.A . 2014 . New Neighborhood Grocery Store Increased Awareness of Food Access but Did Not Alter Dietary Habits or Obesity . Health Affairs 33 ( 2 ): 283 – 91 . Google Scholar CrossRef Search ADS PubMed Dubowitz T. , Ghosh-Dastidar M. , Cohen D.A. , Beckman R. , Steiner E.D. , Hunter G.P. , Florez K.R. , et al. 2015 . Diet and Perceptions Change with Supermarket Introduction in Food Desert, But Not Because of Supermarket Use . Health Affairs 34 ( 11 ): 1858 – 68 . Google Scholar CrossRef Search ADS PubMed Edin K. , Boyd M. , Mabli J. , Ohls J. , Worthington J. , Greene S. , Redel N. , et al. 2013 . SNAP Food Security In-depth Interview Study Final Report . Alexandria VA : U.S. Department of Agriculture, Food and Nutrition Service . Elbel B. , Mijanovich T. , Kiszko K. , Abrams C. , Cantor J. , Dixon L.B . 2017 . The Introduction of a Supermarket via Tax-Credits in a Low-Income Area: The Influence on Purchasing and Consumption . American Journal of Health Promotion 31 ( 1 ): 59 – 66 . Google Scholar CrossRef Search ADS PubMed Gray K.F. , Cunningham K . 2016 . Trends in Supplemental Nutrition Assistance Program Participation Rates: Fiscal Year 2010 to Fiscal Year 2014 . Washington, D.C : Mathematica Policy Research . Grindal T. , Wilde P. , Schwartz G. , Klerman J. , Bartlett S. , Berman D . 2016 . Does Food Retail Access Moderate the Impact of Fruit and Vegetable Incentives for SNAP Participants? Evidence from Western Massachusetts . Food Policy 61 : 59 – 69 . Google Scholar CrossRef Search ADS Handbury J. , Rahkovsky I. , Schnell M . 2016 . Is the Focus on Food Deserts Fruitless? Retail Access and Food Purchases across the Socioeconomic Spectrum. The Wharton School Research Paper No. 91. Available at: https://papers.ssrn.com/sol3/papers.cfm?abstract_id=2757763. Institute of Medicine and National Research Council . 2009 . The Public Health Effects of Food Deserts: Workshop Summary. Washington DC : The National Academies Press . Johnson P. , Giannarelli L. , Huber E. , Betson D . 2015 . National and State-Level Estimates of Special Supplemental Nutrition Program for Women, Infants, and Children (WIC) Eligibles and Program Reach, 2011 . Alexandria VA : U.S. Department of Agriculture Food and Nutrition Service . Kyureghian G. , Nayga R.M. , Bhattacharya S . 2012 . The Effect of Food Store Access and Income on Household Purchases of Fruits and Vegetables: A Mixed Effects Analysis . Applied Economics Policy and Perspectives 35 ( 1 ): 69 – 88 . Google Scholar CrossRef Search ADS 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 ): 74 – 81 . Google Scholar CrossRef Search ADS Mabli J. , Worthington J . 2015 . The Food Access Environment and Food Purchase Behavior of SNAP Households . Journal of Hunger and Environmental Nutrition 10 : 132 – 49 . Google Scholar CrossRef Search ADS Phillips D. , Bell L. , Morgan R. , Pooler J . 2014 . Review of Impact and Examination of Participant Redemption Patterns . Washington DC : U.S. Department of Agriculture . Rose D. , Richards R . 2004 . Food Store Access and Household Fruit and Vegetable Use among Participants in the U.S. Food Stamp Program . Public Health Nutrition 7 ( 8 ): 1081 – 8 . Google Scholar CrossRef Search ADS PubMed Schwartz G. , Grindal T. , Wilde P. , Klerman J. , Bartlett S . 2017 . Supermarket Shopping and The Food Retail Environment among SNAP Participants . Journal of Hunger & Environmental Nutrition 1 – 26 . U.S. Department of Agriculture, Economic Research Service . 2009 . Access to Affordable and Nutritious Food: Measuring and Understanding Food Deserts and Their Consequences. Report to Congress. Administrative Publication No. 36, June 2009. Available at: http://ers.usda.gov/publications/ap-administrative-publication/ap-036.aspx. Castner L. , Henke J . 2011 . Benefit Redemption Patterns in the Supplemental Nutrition Assistance Program . Alexandria VA : U.S. Department of Agriculture Food and Nutrition Service, Office of Research and Analysis . U.S. Department of Agriculture, Food and Nutrition Service . 2016a . Benefit Redemption Patterns in the Supplemental Nutrition Assistance Program. Available at: https://www.fns.usda.gov/snap/benefit-redemption-patterns-supplemental-nutrition-assistance-program. U.S. Department of Agriculture, Food and Nutrition Service . 2016b . WIC Program. Available at: https://www.fns.usda.gov/pd/wic-program. U.S. Department of Agriculture, Food and Nutrition Service . 2017a . Women, Infants and Children (WIC): WIC Eligibility Requirements. Available at: https://www.fns.usda.gov/wic/wic-eligibility-requirements. U.S. Department of Agriculture, Food and Nutrition Service . 2017b . Supplemental Nutrition Assistance Program (SNAP): Eligibility. Available at: https://www.fns.usda.gov/snap/eligibility. Ver Ploeg M. , Mancino L. , Todd J.E. , Clay D.M. , Scharadin B . 2015 . Where Do Americans Usually Shop for Food and How Do They Travel to Get there? Initial Findings from the National Household Food Acquisition and Purchase Survey . Washington DC : U.S. Department of Agriculture, Economic Research Service . ver Ploeg S. , Wilde P . Forthcoming. How Do Food Retail Choices Vary Within and Between Food Retail Environments? Food Policy . Appendix A: Data File Construction This appendix discusses our data sources, geocoding, and sample deletions. A.1 Data Sources Our data come from several sources: SEBTC Sample files, SEBTC EBT records, FNS Retailer listings and the American Community Survey from the U.S. Census Bureau. Each data source and how it is used is described below. Household Data: Our data came from several sources. SEBTC Study Sample data, initially created to verify eligibility for SEBTC and to support stratified random assignment, provided randomization status (i.e., treatment/control), address (for geocoding), and household characteristics (as in table 3). SEBTC EBT data provided information on benefits redeemed. We follow the main SEBTC analysis in computing redemption rates (e.g., Collins et al. 2013). Grantees using the WIC model also provided data on the date, time, and dollar value of each transaction. The data for these sites permit the analysis of redemptions at the food category level. The original issuance data did not include the dollar value of benefits issued. To aggregate across food category levels (e.g., overall redemption rates), the average cost per unit for each food category was imputed, based upon the redemption data. It is thus possible for a household to redeem more than 100% of the benefit (i.e., if costs per unit were higher than average). In practice, such deviations are rare and small. Retailer Data: USDA/FNS provided administrative data on SNAP retailers for each participating state: exact address, store type (which we recoded as supermarket/superstore or not), and participates in WIC (yes/no). We do not have retailer data for neighboring states. Thus, our analysis involves distance to the closest retailer where benefits could be redeemed within the same state. In some cases, the closest retailer to a household may be across the state border, and our analysis cannot account for that. Our main analysis drops households within 1 mile of a state border. Appendix Section A.4 provides sensitivity analyses, including households living within 1 mile of the border. Census Tract Characteristics: To further control for household differences—especially those that might be correlated with distance to a retailer—we augmented the quite limited case-level information with census tract-level information from the American Community Survey (ACS), conducted by the U.S. Census. Geocoding (discussed below) provided census tract for each SEBTC household. ACS samples in any year are too small to generate reliable tract-level estimates. Instead, reported tract-level estimates are centered five-year estimates. Because the study occurred in the summer of 2012, we pulled data from the American Community Survey 2012 5-Year Estimates (available at: https://factfinder.census.gov/faces/nav/jsf/pages/index.xhtml). Population density estimates were drawn from the 2010 Census (available at: https://www.census.gov/quickfacts/meta/long_POP060210.htm and https://www.census.gov/geo/maps-data/data/tiger.html). See table 3 for the specific tract level variables used. A.2 Geocoding We used ArcGIS Online Geocoding Service to geocode (i.e., assign longitude and latitude) to each household and retailer. We then used the Network Analyst tools within ArcGIS to compute the distance from each household to the closest retailer in each class: SNAP/WIC and any/supermarket or superstore. A.3 Sample Deletions Table A.1 describes sample deletions, overall and by site. In 2012, 34,210 households were assigned to the SEBTC treatment group. From that count to the analysis file, we made the following deletions: Ungeocodable: This is an analysis of the impact of distance, where distance is computed via geocoding. We therefore had to drop any records that were not geocoding: no address, P.O. Box, apparently proper address that was not geocodable (see discussion in the next section), distance could not be computed (geocoding software could not connect the address to the road network), and address not in state granting benefits. Tribal Sites: The tribal sites had a high rate of ungeocodable addresses (approximately 12% of all households in tribal sites). The tribal sites were dropped from the main analysis because they may not be representative, considering the high rate of missing addresses. Outliers. Almost all of the households not dropped as ungeocodable were within 10 miles of a retailer (well over 99 percent; see figure A.1). So as not to have outliers drive the regression results, we dropped all households with a distance of greater than 10 miles to closest retailer where they could redeem benefits. Missing Outcomes. Some households could not be matched to EBT records in at least one month. We dropped any household for which we did not have EBT data for every month sufficient to construct the four outcomes considered in this study: redeemed any benefits, redeemed any benefits in a month, percentage of benefits redeemed (unconditional on redeeming any benefits), redeemed all benefits (defined in two ways: as redeeming the maximum amount of benefits in any single month of the demonstration (ever exhausted), and exhausting all benefits over the course of the demonstration (exhausting all)). Appendix B: Sensitivity to Analytic Choices and Additional Substantive Results This appendix provides a sensitivity analysis for analytic choices (Section B.1) and substantive results for additional outcomes (Section B.2). This section also considers sensitivity of the results to sample deletion choices (Section B.3). B.1 Sensitivity to Analytic Choices This section considers sensitivity of the results to analytic choices. Table B.1.1 shows how the key regression results vary with the choice of covariates. Model 1 contains no additional regressors, only the distance measure. Model 2 includes indicators for each of the 14 sites included in the analyses. Model 3 includes the covariates from the sample file in addition to the site-level indicators. Model 4 includes the sample covariates, the site-level indicators, and the census tract characteristics from the American Community Survey. When no additional regressors are included, there is a significant, negative relationship between the proportion of benefits redeemed and the distance to the nearest supermarket where benefits could be redeemed. For each additional mile farther between a household and a supermarket, a participant would redeem 2% less of his or her benefits. A household that was a maximum of 10 miles away would redeem 20% less of its benefits. For SNAP households alone, the relationship is much smaller, and positive. The relationship seems to be driven by the WIC sites. However, as soon as we control for any additional covariates, the relationship disappears. The relationship between distance and redemption rates appears to be spurious. We ran several variations of our analytic model on the focal outcome (unconditional percentage of benefits redeemed), and additional model specifications yielded the same results. First, we looked at the relationship between distance to the nearest retailer of any type where a participant could redeem benefits (as opposed to closest supermarket). There seems to be no relationship between distance to the closest retailer where participants could redeem benefits and the proportion of their benefits they actually redeem. See table B.1.2. We also examined the relationship between distance to the nearest supermarket where participants could redeem benefits, and redemption patterns, controlling for the distance to the nearest store of any type where participants could redeem benefits. Results are shown in table B.1.3. The pooled results show no relationship between distance to either store type or proportion of benefits redeemed. In SNAP sites, when controlling for distance to the store of any type, there was a small, negative relationship between distance to the nearest supermarket and proportion of benefits redeemed, although the relationship is insignificant and quite close to 0. Lastly, we controlled for a non-linear relationship between distance and redemption patterns, shown in table B.1.4. Controlling for a spline at 1 mile, there is no significant relationship between distance to the nearest supermarket and redemption rates. However, in SNAP sites, a participant who is more than 10 miles away from the nearest retailer would redeem approximately 10% more of his or her benefits. The pattern does not hold when the spline is at ¾ of a mile (table B.1.4) B.2 Substantive Results for Additional Outcomes We also conducted analyses on three additional outcomes: whether a household redeemed any benefit, whether a household exhausted its benefits in any month of the SEBTC demonstration, and whether a household exhausted all of its benefits over the course of the demonstration. Table B.2.1 shows the results for participants redeeming any of their benefits, proportion redeemed (conditional on redeeming any benefit), exhausted benefits in any month, and exhausted benefits in all months. These models include the full set of covariates. The patterns for redeeming any benefit are similar to the patterns for proportion of benefits redeemed. There is no detectable relationship between distance and redemption, exhausting benefits in a given month, or proportion of benefits redeemed (conditional on redeeming any benefit). However, for SNAP households, there is a negative, significant relationship between distance to the nearest supermarket where participants could redeem benefits and their likelihood of exhausting all of their benefits over the course of the SEBTC demonstration. A household that is 10 miles farther from a supermarket would be 7% less likely to redeem all of its benefits. B.3 Sensitivity to Sample Selection This section considers sensitivity of the results to sample deletion choices. The analysis is impossible without addresses, so an analysis of the impact of losing records to failure to geocode is not possible. Table B.3.1 presents the main results for: (a) the analysis sample; (b) adding those within 1 mile of a state border; and (c) adding the two tribal sites (but still excluding those within 1 mile of a state border). The results are largely consistent for each sample. When the cases close to state borders are included, there are no changes to the coefficients. When the tribal sites are included, the coefficient for WIC is slightly larger, but still insignificant. Table A.1.1 Sample Attrition by Site Total Households Missing Address PO Box Could Not Be Geocoded No Distance Calculation Out of State Outlier Distance (>10 miles) < 1 Mile to State Border Missing EBT Data Analysis Sample Total Households 34,520 1 782 11 45 10 115 733 2,524 30,299 All SNAP Sites 17,738 1 98 9 31 1 11 727 283 16,577 Eastern Connecticut 2,276 1 11 0 0 1 6 23 70 2,164 Western Connecticut 1,286 0 1 0 0 0 0 0 28 1,257 Delaware 2,870 0 40 0 9 0 0 0 3 2,818 Kansas City, MO 2,970 0 0 1 1 0 0 123 79 2,766 St. Louis, MO 3,468 0 1 0 20 0 0 488 9 2,950 Deschutes, OR 1,728 0 7 8 1 0 5 0 90 1,617 Salem, OR 1,573 0 14 0 0 0 0 0 0 1,559 Washington 1,567 0 24 0 0 0 0 93 4 1,446 All WIC Sites 16,782 0 684 2 14 9 104 6 2,241 13,722 Cherokee 2,748 0 589 0 1 5 18 3 375 1,757 Chickasaw 2,559 0 63 1 0 0 58 1 464 1,972 Eastern Michigan 2,630 0 24 0 1 0 0 0 1,183 1,422 Grand Rapids, MI 2,782 0 1 1 0 0 0 0 7 2,773 Nevada 3,049 0 7 0 12 4 25 2 112 2,887 Texas 3,014 0 0 0 0 0 3 0 100 2,911 Total Households Missing Address PO Box Could Not Be Geocoded No Distance Calculation Out of State Outlier Distance (>10 miles) < 1 Mile to State Border Missing EBT Data Analysis Sample Total Households 34,520 1 782 11 45 10 115 733 2,524 30,299 All SNAP Sites 17,738 1 98 9 31 1 11 727 283 16,577 Eastern Connecticut 2,276 1 11 0 0 1 6 23 70 2,164 Western Connecticut 1,286 0 1 0 0 0 0 0 28 1,257 Delaware 2,870 0 40 0 9 0 0 0 3 2,818 Kansas City, MO 2,970 0 0 1 1 0 0 123 79 2,766 St. Louis, MO 3,468 0 1 0 20 0 0 488 9 2,950 Deschutes, OR 1,728 0 7 8 1 0 5 0 90 1,617 Salem, OR 1,573 0 14 0 0 0 0 0 0 1,559 Washington 1,567 0 24 0 0 0 0 93 4 1,446 All WIC Sites 16,782 0 684 2 14 9 104 6 2,241 13,722 Cherokee 2,748 0 589 0 1 5 18 3 375 1,757 Chickasaw 2,559 0 63 1 0 0 58 1 464 1,972 Eastern Michigan 2,630 0 24 0 1 0 0 0 1,183 1,422 Grand Rapids, MI 2,782 0 1 1 0 0 0 0 7 2,773 Nevada 3,049 0 7 0 12 4 25 2 112 2,887 Texas 3,014 0 0 0 0 0 3 0 100 2,911 Table A.1.1 Sample Attrition by Site Total Households Missing Address PO Box Could Not Be Geocoded No Distance Calculation Out of State Outlier Distance (>10 miles) < 1 Mile to State Border Missing EBT Data Analysis Sample Total Households 34,520 1 782 11 45 10 115 733 2,524 30,299 All SNAP Sites 17,738 1 98 9 31 1 11 727 283 16,577 Eastern Connecticut 2,276 1 11 0 0 1 6 23 70 2,164 Western Connecticut 1,286 0 1 0 0 0 0 0 28 1,257 Delaware 2,870 0 40 0 9 0 0 0 3 2,818 Kansas City, MO 2,970 0 0 1 1 0 0 123 79 2,766 St. Louis, MO 3,468 0 1 0 20 0 0 488 9 2,950 Deschutes, OR 1,728 0 7 8 1 0 5 0 90 1,617 Salem, OR 1,573 0 14 0 0 0 0 0 0 1,559 Washington 1,567 0 24 0 0 0 0 93 4 1,446 All WIC Sites 16,782 0 684 2 14 9 104 6 2,241 13,722 Cherokee 2,748 0 589 0 1 5 18 3 375 1,757 Chickasaw 2,559 0 63 1 0 0 58 1 464 1,972 Eastern Michigan 2,630 0 24 0 1 0 0 0 1,183 1,422 Grand Rapids, MI 2,782 0 1 1 0 0 0 0 7 2,773 Nevada 3,049 0 7 0 12 4 25 2 112 2,887 Texas 3,014 0 0 0 0 0 3 0 100 2,911 Total Households Missing Address PO Box Could Not Be Geocoded No Distance Calculation Out of State Outlier Distance (>10 miles) < 1 Mile to State Border Missing EBT Data Analysis Sample Total Households 34,520 1 782 11 45 10 115 733 2,524 30,299 All SNAP Sites 17,738 1 98 9 31 1 11 727 283 16,577 Eastern Connecticut 2,276 1 11 0 0 1 6 23 70 2,164 Western Connecticut 1,286 0 1 0 0 0 0 0 28 1,257 Delaware 2,870 0 40 0 9 0 0 0 3 2,818 Kansas City, MO 2,970 0 0 1 1 0 0 123 79 2,766 St. Louis, MO 3,468 0 1 0 20 0 0 488 9 2,950 Deschutes, OR 1,728 0 7 8 1 0 5 0 90 1,617 Salem, OR 1,573 0 14 0 0 0 0 0 0 1,559 Washington 1,567 0 24 0 0 0 0 93 4 1,446 All WIC Sites 16,782 0 684 2 14 9 104 6 2,241 13,722 Cherokee 2,748 0 589 0 1 5 18 3 375 1,757 Chickasaw 2,559 0 63 1 0 0 58 1 464 1,972 Eastern Michigan 2,630 0 24 0 1 0 0 0 1,183 1,422 Grand Rapids, MI 2,782 0 1 1 0 0 0 0 7 2,773 Nevada 3,049 0 7 0 12 4 25 2 112 2,887 Texas 3,014 0 0 0 0 0 3 0 100 2,911 Figure A.1 View largeDownload slide Percent of benefits redeemed by distance to the nearest supermarket of superstore Figure A.1 View largeDownload slide Percent of benefits redeemed by distance to the nearest supermarket of superstore Table B.1.1 Effects of Distance to the Closest Supermarket on Percentage Redeemed for Multiple Regression Models Model 1 Model 2 Model 3 Model 4 SNAP sites, 8 sites Driving distance to the nearest supermarket where benefits could be redeemed 0.002* −0.000 −0.000 −0.001 (0.001) (0.001) (0.001) (0.001) Constant 0.936*** 0.959*** 0.909*** 0.878*** (0.002) (0.004) (0.060) (0.066) R-squared 0.000 0.016 0.025 0.026 N 16,577 16,577 16,577 16,577 WIC sites, 4 sites Driving distance to the nearest supermarket where benefits could be redeemed −0.017*** 0.001 0.001 0.001 (0.002) (0.002) (0.002) (0.002) Constant 0.604*** 0.641*** 0.526*** 0.680*** (0.004) (0.008) (0.061) (0.074) R-squared 0.007 0.052 0.070 0.075 N 9,993 9,993 9,993 9,993 Model 1 Model 2 Model 3 Model 4 SNAP sites, 8 sites Driving distance to the nearest supermarket where benefits could be redeemed 0.002* −0.000 −0.000 −0.001 (0.001) (0.001) (0.001) (0.001) Constant 0.936*** 0.959*** 0.909*** 0.878*** (0.002) (0.004) (0.060) (0.066) R-squared 0.000 0.016 0.025 0.026 N 16,577 16,577 16,577 16,577 WIC sites, 4 sites Driving distance to the nearest supermarket where benefits could be redeemed −0.017*** 0.001 0.001 0.001 (0.002) (0.002) (0.002) (0.002) Constant 0.604*** 0.641*** 0.526*** 0.680*** (0.004) (0.008) (0.061) (0.074) R-squared 0.007 0.052 0.070 0.075 N 9,993 9,993 9,993 9,993 Note: Model 1: no regressors; Model 2: site indicators only; Model 3: site indicators and household characteristics; Model 4: site indicators, household characteristics for the household itself and household characteristics for the census tract (computed from 5-year ACS tabulations). Two-sided hypothesis test; asterisks *** = statistically significant, p<0.01; ** = statistically significant, p<0.05; * = statistically significant, p<0.10. View Large Table B.1.1 Effects of Distance to the Closest Supermarket on Percentage Redeemed for Multiple Regression Models Model 1 Model 2 Model 3 Model 4 SNAP sites, 8 sites Driving distance to the nearest supermarket where benefits could be redeemed 0.002* −0.000 −0.000 −0.001 (0.001) (0.001) (0.001) (0.001) Constant 0.936*** 0.959*** 0.909*** 0.878*** (0.002) (0.004) (0.060) (0.066) R-squared 0.000 0.016 0.025 0.026 N 16,577 16,577 16,577 16,577 WIC sites, 4 sites Driving distance to the nearest supermarket where benefits could be redeemed −0.017*** 0.001 0.001 0.001 (0.002) (0.002) (0.002) (0.002) Constant 0.604*** 0.641*** 0.526*** 0.680*** (0.004) (0.008) (0.061) (0.074) R-squared 0.007 0.052 0.070 0.075 N 9,993 9,993 9,993 9,993 Model 1 Model 2 Model 3 Model 4 SNAP sites, 8 sites Driving distance to the nearest supermarket where benefits could be redeemed 0.002* −0.000 −0.000 −0.001 (0.001) (0.001) (0.001) (0.001) Constant 0.936*** 0.959*** 0.909*** 0.878*** (0.002) (0.004) (0.060) (0.066) R-squared 0.000 0.016 0.025 0.026 N 16,577 16,577 16,577 16,577 WIC sites, 4 sites Driving distance to the nearest supermarket where benefits could be redeemed −0.017*** 0.001 0.001 0.001 (0.002) (0.002) (0.002) (0.002) Constant 0.604*** 0.641*** 0.526*** 0.680*** (0.004) (0.008) (0.061) (0.074) R-squared 0.007 0.052 0.070 0.075 N 9,993 9,993 9,993 9,993 Note: Model 1: no regressors; Model 2: site indicators only; Model 3: site indicators and household characteristics; Model 4: site indicators, household characteristics for the household itself and household characteristics for the census tract (computed from 5-year ACS tabulations). Two-sided hypothesis test; asterisks *** = statistically significant, p<0.01; ** = statistically significant, p<0.05; * = statistically significant, p<0.10. View Large Table B.1.2 Effects of Distance to the Closest Retailer on Percentage Redeemed SNAP Households WIC Households Driving distance to the nearest retailer where benefits could be redeemed 0.000 −0.000 (0.002) (0.002) Constant 0.877*** 0.683*** (0.066) (0.074) R-squared 0.026 0.075 N 16,577 9,993 SNAP Households WIC Households Driving distance to the nearest retailer where benefits could be redeemed 0.000 −0.000 (0.002) (0.002) Constant 0.877*** 0.683*** (0.066) (0.074) R-squared 0.026 0.075 N 16,577 9,993 Note: Two-sided hypothesis test; asterisks *** = statistically significant, p<0.01; ** = statistically significant, p<0.05; * = statistically significant, p<0.10. View Large Table B.1.2 Effects of Distance to the Closest Retailer on Percentage Redeemed SNAP Households WIC Households Driving distance to the nearest retailer where benefits could be redeemed 0.000 −0.000 (0.002) (0.002) Constant 0.877*** 0.683*** (0.066) (0.074) R-squared 0.026 0.075 N 16,577 9,993 SNAP Households WIC Households Driving distance to the nearest retailer where benefits could be redeemed 0.000 −0.000 (0.002) (0.002) Constant 0.877*** 0.683*** (0.066) (0.074) R-squared 0.026 0.075 N 16,577 9,993 Note: Two-sided hypothesis test; asterisks *** = statistically significant, p<0.01; ** = statistically significant, p<0.05; * = statistically significant, p<0.10. View Large Table B.1.3 Effects of Multiple Distances on Percentage Redeemed All Households SNAP Households WIC Households Driving distance to the nearest supermarket where benefits could be redeemed 0.000 −0.002 0.004 (0.002) (0.002) (0.003) Driving distance to the nearest retailer of any type 0.000 0.002 −0.004 (0.002) (0.003) (0.004) Constant 0.605*** 0.878*** 0.680*** (0.050) (0.066) (0.074) R-squared 0.345 0.026 0.075 N 26,570 16,577 9,993 All Households SNAP Households WIC Households Driving distance to the nearest supermarket where benefits could be redeemed 0.000 −0.002 0.004 (0.002) (0.002) (0.003) Driving distance to the nearest retailer of any type 0.000 0.002 −0.004 (0.002) (0.003) (0.004) Constant 0.605*** 0.878*** 0.680*** (0.050) (0.066) (0.074) R-squared 0.345 0.026 0.075 N 26,570 16,577 9,993 Note: Two-sided hypothesis test; asterisks *** = statistically significant, p<0.01; ** = statistically significant, p<0.05; * = statistically significant, p<0.10. View Large Table B.1.3 Effects of Multiple Distances on Percentage Redeemed All Households SNAP Households WIC Households Driving distance to the nearest supermarket where benefits could be redeemed 0.000 −0.002 0.004 (0.002) (0.002) (0.003) Driving distance to the nearest retailer of any type 0.000 0.002 −0.004 (0.002) (0.003) (0.004) Constant 0.605*** 0.878*** 0.680*** (0.050) (0.066) (0.074) R-squared 0.345 0.026 0.075 N 26,570 16,577 9,993 All Households SNAP Households WIC Households Driving distance to the nearest supermarket where benefits could be redeemed 0.000 −0.002 0.004 (0.002) (0.002) (0.003) Driving distance to the nearest retailer of any type 0.000 0.002 −0.004 (0.002) (0.003) (0.004) Constant 0.605*** 0.878*** 0.680*** (0.050) (0.066) (0.074) R-squared 0.345 0.026 0.075 N 26,570 16,577 9,993 Note: Two-sided hypothesis test; asterisks *** = statistically significant, p<0.01; ** = statistically significant, p<0.05; * = statistically significant, p<0.10. View Large Table B.1.4 Effects of Distance to the Closest Supermarket on Percentage Redeemed with Spline at 3/4 Mile SNAP Households WIC Households 0.75 Miles 1.0 Mile 0.75 Miles 1.0 Mile Driving distance to the nearest supermarket where benefits could be redeemed −0.000 0.000 0.001 0.000 (0.001) (0.001) (0.002) (0.002) Spline = 1 Mile 0.007 0.009* 0.007 −0.005 (0.007) (0.005) (0.015) (0.011) Constant 0.875*** 0.871*** 0.678*** 0.682*** (0.066) (0.066) (0.075) (0.075) R-squared 0.026 0.026 0.075 0.075 N 16,577 16,577 9,993 9,993 SNAP Households WIC Households 0.75 Miles 1.0 Mile 0.75 Miles 1.0 Mile Driving distance to the nearest supermarket where benefits could be redeemed −0.000 0.000 0.001 0.000 (0.001) (0.001) (0.002) (0.002) Spline = 1 Mile 0.007 0.009* 0.007 −0.005 (0.007) (0.005) (0.015) (0.011) Constant 0.875*** 0.871*** 0.678*** 0.682*** (0.066) (0.066) (0.075) (0.075) R-squared 0.026 0.026 0.075 0.075 N 16,577 16,577 9,993 9,993 Note: Two-sided hypothesis test; asterisks *** = statistically significant, p<0.01; ** = statistically significant, p<0.05; * = statistically significant, p<0.10. View Large Table B.1.4 Effects of Distance to the Closest Supermarket on Percentage Redeemed with Spline at 3/4 Mile SNAP Households WIC Households 0.75 Miles 1.0 Mile 0.75 Miles 1.0 Mile Driving distance to the nearest supermarket where benefits could be redeemed −0.000 0.000 0.001 0.000 (0.001) (0.001) (0.002) (0.002) Spline = 1 Mile 0.007 0.009* 0.007 −0.005 (0.007) (0.005) (0.015) (0.011) Constant 0.875*** 0.871*** 0.678*** 0.682*** (0.066) (0.066) (0.075) (0.075) R-squared 0.026 0.026 0.075 0.075 N 16,577 16,577 9,993 9,993 SNAP Households WIC Households 0.75 Miles 1.0 Mile 0.75 Miles 1.0 Mile Driving distance to the nearest supermarket where benefits could be redeemed −0.000 0.000 0.001 0.000 (0.001) (0.001) (0.002) (0.002) Spline = 1 Mile 0.007 0.009* 0.007 −0.005 (0.007) (0.005) (0.015) (0.011) Constant 0.875*** 0.871*** 0.678*** 0.682*** (0.066) (0.066) (0.075) (0.075) R-squared 0.026 0.026 0.075 0.075 N 16,577 16,577 9,993 9,993 Note: Two-sided hypothesis test; asterisks *** = statistically significant, p<0.01; ** = statistically significant, p<0.05; * = statistically significant, p<0.10. View Large Table B.2.1 Regression Results for Alternate Outcomes Redeemed Any Benefit Conditional Proportion Redeemed Exhausted Benefits in Any Month Exhausted Benefits in All Months SNAP sites, 8 sites Driving distance to the nearest supermarket where benefits could be redeemed −0.001 −0.000 −0.003 −0.007** (0.001) (0.001) (0.003) (0.003) Constant 0.950*** 0.924*** 0.854*** −0.007 (0.057) (0.042) (0.077) (0.096) R-squared 0.016 0.016 0.067 0.127 N 16,577 15,923 16,577 16,577 WIC sites, 8 sites Driving distance to the nearest supermarket where benefits could be redeemed 0.001 0.000 0.001 0.000 (0.002) (0.001) (0.001) (0.000) Constant 0.654*** 0.997*** 0.551*** 0.024 (0.086) (0.047) (0.074) (0.017) R-squared 0.063 0.314 0.172 0.014 N 9,993 8,359 9,993 9,993 Redeemed Any Benefit Conditional Proportion Redeemed Exhausted Benefits in Any Month Exhausted Benefits in All Months SNAP sites, 8 sites Driving distance to the nearest supermarket where benefits could be redeemed −0.001 −0.000 −0.003 −0.007** (0.001) (0.001) (0.003) (0.003) Constant 0.950*** 0.924*** 0.854*** −0.007 (0.057) (0.042) (0.077) (0.096) R-squared 0.016 0.016 0.067 0.127 N 16,577 15,923 16,577 16,577 WIC sites, 8 sites Driving distance to the nearest supermarket where benefits could be redeemed 0.001 0.000 0.001 0.000 (0.002) (0.001) (0.001) (0.000) Constant 0.654*** 0.997*** 0.551*** 0.024 (0.086) (0.047) (0.074) (0.017) R-squared 0.063 0.314 0.172 0.014 N 9,993 8,359 9,993 9,993 Note: Two-sided hypothesis test; asterisks *** = statistically significant, p<0.01; ** = statistically significant, p<0.05; * = statistically significant, p<0.10. View Large Table B.2.1 Regression Results for Alternate Outcomes Redeemed Any Benefit Conditional Proportion Redeemed Exhausted Benefits in Any Month Exhausted Benefits in All Months SNAP sites, 8 sites Driving distance to the nearest supermarket where benefits could be redeemed −0.001 −0.000 −0.003 −0.007** (0.001) (0.001) (0.003) (0.003) Constant 0.950*** 0.924*** 0.854*** −0.007 (0.057) (0.042) (0.077) (0.096) R-squared 0.016 0.016 0.067 0.127 N 16,577 15,923 16,577 16,577 WIC sites, 8 sites Driving distance to the nearest supermarket where benefits could be redeemed 0.001 0.000 0.001 0.000 (0.002) (0.001) (0.001) (0.000) Constant 0.654*** 0.997*** 0.551*** 0.024 (0.086) (0.047) (0.074) (0.017) R-squared 0.063 0.314 0.172 0.014 N 9,993 8,359 9,993 9,993 Redeemed Any Benefit Conditional Proportion Redeemed Exhausted Benefits in Any Month Exhausted Benefits in All Months SNAP sites, 8 sites Driving distance to the nearest supermarket where benefits could be redeemed −0.001 −0.000 −0.003 −0.007** (0.001) (0.001) (0.003) (0.003) Constant 0.950*** 0.924*** 0.854*** −0.007 (0.057) (0.042) (0.077) (0.096) R-squared 0.016 0.016 0.067 0.127 N 16,577 15,923 16,577 16,577 WIC sites, 8 sites Driving distance to the nearest supermarket where benefits could be redeemed 0.001 0.000 0.001 0.000 (0.002) (0.001) (0.001) (0.000) Constant 0.654*** 0.997*** 0.551*** 0.024 (0.086) (0.047) (0.074) (0.017) R-squared 0.063 0.314 0.172 0.014 N 9,993 8,359 9,993 9,993 Note: Two-sided hypothesis test; asterisks *** = statistically significant, p<0.01; ** = statistically significant, p<0.05; * = statistically significant, p<0.10. View Large Table B.3.1 Sensitivity to Sample Deletion Analytic Model Including Household Near State Borders Including Tribal Sites SNAP sites, 8 sites Driving distance to the nearest supermarket where benefits could be redeemed −0.001 −0.000 −0.001 (0.001) (0.000) (0.001) Constant 0.878*** 0.876*** 0.878*** (0.066) (0.021) (0.066) R-squared 0.026 0.027 0.026 N 16,577 17,304 16,577 WIC sites, 4 sites Driving distance to the nearest supermarket where benefits could be redeemed 0.001 0.001 0.003 (0.002) (0.001) (0.001) Constant 0.680*** 0.560*** 0.546*** (0.074) (0.076) (0.071) R-squared 0.075 0.075 0.083 N 9,993 10,006 13,772 Analytic Model Including Household Near State Borders Including Tribal Sites SNAP sites, 8 sites Driving distance to the nearest supermarket where benefits could be redeemed −0.001 −0.000 −0.001 (0.001) (0.000) (0.001) Constant 0.878*** 0.876*** 0.878*** (0.066) (0.021) (0.066) R-squared 0.026 0.027 0.026 N 16,577 17,304 16,577 WIC sites, 4 sites Driving distance to the nearest supermarket where benefits could be redeemed 0.001 0.001 0.003 (0.002) (0.001) (0.001) Constant 0.680*** 0.560*** 0.546*** (0.074) (0.076) (0.071) R-squared 0.075 0.075 0.083 N 9,993 10,006 13,772 Note: Two-sided hypothesis test; asterisks *** = statistically significant, p<0.01; ** = statistically significant, p<0.05; * = statistically significant, p<0.10. View Large Table B.3.1 Sensitivity to Sample Deletion Analytic Model Including Household Near State Borders Including Tribal Sites SNAP sites, 8 sites Driving distance to the nearest supermarket where benefits could be redeemed −0.001 −0.000 −0.001 (0.001) (0.000) (0.001) Constant 0.878*** 0.876*** 0.878*** (0.066) (0.021) (0.066) R-squared 0.026 0.027 0.026 N 16,577 17,304 16,577 WIC sites, 4 sites Driving distance to the nearest supermarket where benefits could be redeemed 0.001 0.001 0.003 (0.002) (0.001) (0.001) Constant 0.680*** 0.560*** 0.546*** (0.074) (0.076) (0.071) R-squared 0.075 0.075 0.083 N 9,993 10,006 13,772 Analytic Model Including Household Near State Borders Including Tribal Sites SNAP sites, 8 sites Driving distance to the nearest supermarket where benefits could be redeemed −0.001 −0.000 −0.001 (0.001) (0.000) (0.001) Constant 0.878*** 0.876*** 0.878*** (0.066) (0.021) (0.066) R-squared 0.026 0.027 0.026 N 16,577 17,304 16,577 WIC sites, 4 sites Driving distance to the nearest supermarket where benefits could be redeemed 0.001 0.001 0.003 (0.002) (0.001) (0.001) Constant 0.680*** 0.560*** 0.546*** (0.074) (0.076) (0.071) R-squared 0.075 0.075 0.083 N 9,993 10,006 13,772 Note: Two-sided hypothesis test; asterisks *** = statistically significant, p<0.01; ** = statistically significant, p<0.05; * = statistically significant, p<0.10. View Large © The Author(s) 2018. Published by Oxford University Press on behalf of the Agricultural and Applied Economics Association. All rights reserved. For Permissions, please e-mail: journals.permissions@oup.com

Journal

Applied Economic Perspectives and PolicyOxford University Press

Published: Feb 28, 2018

There are no references for this article.

You’re reading a free preview. Subscribe to read the entire article.


DeepDyve is your
personal research library

It’s your single place to instantly
discover and read the research
that matters to you.

Enjoy affordable access to
over 18 million articles from more than
15,000 peer-reviewed journals.

All for just $49/month

Explore the DeepDyve Library

Search

Query the DeepDyve database, plus search all of PubMed and Google Scholar seamlessly

Organize

Save any article or search result from DeepDyve, PubMed, and Google Scholar... all in one place.

Access

Get unlimited, online access to over 18 million full-text articles from more than 15,000 scientific journals.

Your journals are on DeepDyve

Read from thousands of the leading scholarly journals from SpringerNature, Elsevier, Wiley-Blackwell, Oxford University Press and more.

All the latest content is available, no embargo periods.

See the journals in your area

DeepDyve

Freelancer

DeepDyve

Pro

Price

FREE

$49/month
$360/year

Save searches from
Google Scholar,
PubMed

Create lists to
organize your research

Export lists, citations

Read DeepDyve articles

Abstract access only

Unlimited access to over
18 million full-text articles

Print

20 pages / month

PDF Discount

20% off