In the past decade, a number of studies have examined whether residents of socioeconomically disadvantaged neighborhoods are more likely to be obese than those of more affluent neighborhoods [1
]. These studies have relied on geographic information systems (GIS) to characterize neighborhood characteristics combined with information on residents’ health status. This approach was further expanded to include characterization of the neighborhoods with respect to food availability and accessibility characteristics [6
], which ultimately led to the landmark report by the United States (US) Department of Agriculture’s Economic Research Service on “Access to Affordable and Nutritious Food: Measuring and Understanding Food Deserts and Their Consequences” [8
] and the coining of the term “food deserts” [9
A key assumption of this line of research has been that people actually shop in the neighborhoods in which they live. However, several studies have now shown this is not the case, as many shoppers do not select the nearest grocery store [10
]. For example, in a study of residents of the South Carolina Midlands region, we found that the utilized grocery stores were on average 3.2 miles farther away than the nearest stores for residents of food deserts and 4.2 miles farther away for residents of non-deserts [10
]. It is therefore not surprising that, in that same population, the characteristics of the built environment (e.g., availability of stores in the neighborhood) were not directly associated with fruit and vegetable intake (they were indirectly associated) and that, instead, actual food shopping behaviors (e.g., frequency of grocery shopping) were the central characteristics associated with fruit and vegetable intake [17
]. Another hypothesis tested by previous research has been that shopping at supermarkets or large grocery stores is associated with healthier grocery selection, healthier eating behavior, and lower body mass index (BMI) [18
], and some studies have indeed shown store type to be associated with quality of dietary intake [15
]; however, no studies focusing on BMI have considered the actual types of foods being purchased.
Very few studies have collected detailed information on food shopping behaviors, including characteristics of the stores utilized (e.g., type, price level, distance, etc.) and mode of transportation, and have been able to relate these to obesity [11
]. Although findings have been mixed with respect to which food shopping behaviors are associated with obesity, there seems to be some consistency, with studies finding few associations between obesity and distance to utilized store [11
] but relatively consistent evidence for certain store types—particularly discount grocery stores—being associated with higher BMI and obesity [11
]. However, most studies to date have focused their assessment of food shopping behaviors on the primary store, even though most households tend to utilize multiple stores [25
Thus, the purpose of this analysis was to comprehensively evaluate the association of food shopping at the three most frequented grocery stores and food acquisition behaviors (e.g., obtaining food at churches or social service organizations, food banks or pantries) with BMI among residents of two low-income communities in South Carolina, 81% of whom lived in food desert census tracts, defined as a tract with a low-income population having low access to a supermarket or supercenter [8
]. We hypothesized that distance to and type of grocery store utilized; frequency of shopping; acquiring food at food banks or pantries or from a church or other social service organization; perceptions of the built food environment; and food environment access characteristics would be significant predictors of BMI. These hypotheses were based on a conceptual framework we developed previously [17
Participants were approximately equally distributed across the two study locations in South Carolina (47% vs. 53%), and 89% of participants in Location 1 and 74% in Location 2 resided in food desert census tracts.
Characteristics of the study population with complete data for this analysis are summarized in Table 1
. Most of the participants were African American (93%) and female (80%); had an average age of 52 years; and only about 13% were married or living with a partner. A large percentage were food insecure (62%), participated in SNAP (66%), had completed at most high school or a lower level of education (69%), and had an annual household income less than $
20,000 (79%). The average BMI was 32.5 kg/m2
, and 55% of the sample was obese.
additionally shows food shopping characteristics of participants’ three most-frequented stores. On average, participants traveled 2.5 miles to their primarily utilized store and slightly farther to stores 2 and 3. Only 45% of the sample used a vehicle of their own to get to store 1; 36% rode in someone else’s vehicle, 8.9% took a bus or taxi, and 10% walked or used a bicycle. Supermarkets were the most frequented type of store utilized (61% for store 1, 64% for store 2, and 67% for store 3), followed by supercenters (27%, 22%, and 14%, respectively). Only a small percentage (i.e., 12%, 14%, and 19%, respectively) selected another store type (e.g., convenience, drug/pharmacy, dollar, specialty, or smaller grocery store) for stores 1, 2, and 3. Shopping frequency varied substantially across the utilized stores, with about 60% of participants shopping less than once a week at their primary store, 34% shopping less than twice a month at store 2, and 66% shopping less than once a month at store 3.
The majority of participants indicated that they additionally acquired food at food banks or food pantries (53%) or from a church or social service organization (54%), and less than half (46%) shopped at a farmers’ market in the previous year. The nearest supermarket/large grocery store was on average 1.5 miles away, and the nearest supercenter/warehouse club was 2.7 miles away.
Participants largely perceived their food environment as not supporting healthy food availability (average 4.7 on a scale ranging from 0 to 12), on average considered lack of access to adequate food shopping a minor or a somewhat serious problem (average 1.5 on a scale ranging from 0 to 3), and were relatively neutral in terms of having opportunities to purchase fast foods (average 1.8 on a scale ranging from 0 to 4).
presents results from the hierarchical linear models. Shopping at a supercenter/warehouse club for store 1 compared to shopping at a supermarket was positively and significantly associated with BMI in Model 1, and shopping at an “other” type of store for store 3 was inversely associated with BMI, after controlling for demographic and socioeconomic covariates. Acquiring food at food banks/pantries or churches/social services or a farmers’ market was not associated with BMI. These findings remained unchanged when participant perceptions of the food environment were added to the model (Model 2). Expressing the perception that lack of access to adequate food shopping in the neighborhood was not really a problem was inversely associated with BMI, i.e., the less a person considered lack of access a problem, the lower his/her BMI. Perceptions of healthy food availability or fast food availability were not significantly associated with BMI. The correlation between the lack of access question (i.e., access not really being a problem) and the availability of healthy foods in the neighborhood score was significant but moderate in size (r = 0.37). Further consideration of the minimum distance a participant would have to travel to reach the nearest supermarket or supercenter/warehouse club and whether the participant resided in a food desert census tract (Model 3) did not change the findings. Food desert status of the census tract of residence was not significantly associated with BMI. Additionally, the model R2
increased across the models from 0.15754 in Model 1 to 0.1730 and 0.1768 in Models 2 and 3, respectively, with the increase in R2
being statistically significant. The small magnitude of the squared semi-partial correlations indicated that no one single predictor variable explained a substantial amount of the variation.
We additionally estimated the mean BMI levels for participants and differences between groups of participants from Model 3. Adjusted for all covariates, those shopping at supercenters/warehouse clubs for store 1 had a 2.6 kg/m2 higher BMI than those who shopped at supermarkets (31.6 kg/m2 vs. 28.9 kg/m2, respectively), and participants who shopped at an “other” type of store for store 3 had a 2.6 kg/m2 lower BMI than those who shopped at a supermarket for store 3 (27.3 kg/m2 vs. 29.9 kg/m2, respectively). In addition, a one-unit higher response on the lack of food access question (i.e., indicative of food access not being a problem) was associated with a 0.86 kg/m2 lower BMI.
The present study focused on very low income populations in two Southern cities, the majority of whom were of minority race/ethnicity and resided in census tracts designated by the USDA as food desert tracts at the study’s inception. Ours was a population experiencing very high poverty, with 46% of shoppers reporting incomes below $
10,000 and more than half acquiring food at food banks, food pantries, churches, and social service organizations. A similar population was recently studied in Pittsburg, Pennsylvania [15
]. The majority of our study’s participants shopped at grocery stores inside the two geographic study areas, as defined by the ground-truthing survey (62–65% for stores 1–3). However, the utilized grocery stores within the study areas were all located in the non-food desert census tracts, where only 11% (in Location 1) and 26% (in Location 2) of the populations lived. Similar findings were reported by LeDoux and Vojnovic (2013), who demonstrated that residents of census tracts in the lower east side of Detroit, Michigan, generally did not utilize their immediate food environment for grocery shopping and selectively shopped elsewhere [14
This study is one of a relatively small number that have focused explicitly on actual grocery shopping behaviors characterized by GIS measures to utilized stores and their relation to BMI. Consistent with several previous reports, we did not find evidence for an association of distance to the primarily utilized grocery store and BMI [11
]. However, we additionally offer evidence that distance to the second or third most-frequented grocery store is also not associated with BMI. This suggests that previous null findings on distance and BMI in studies assessing only a single store are not likely to have been confounded by lack of information on other utilized stores, i.e., by potential trade-offs made by consumers with respect to time and distance to multiple utilized stores. Moreover, comparing the distance to each of the three stores suggested that participants chose a primary store that was on average slightly more proximal (0.2 miles) than their second most-frequented store, which was in turn 0.2 miles closer than the third store. All three stores were, on average, farther away than the nearest supermarket, which suggests that these shoppers do not in general shop at the most proximal location [10
Contrary to our hypothesis, which had been informed by previous work [15
], shopping frequency was not associated with BMI. Most studies to date have not included data on shopping frequency at utilized stores [11
], and hence there are few data to which we can compare our findings. The average shopping frequency at the primary grocery store of 1.2 times per week found here is lower than the 1.8 times per week we reported in a different area of South Carolina [10
]. In a study among SNAP recipients in eastern North Carolina, Jillcott-Pitts et al. (2012) demonstrated that the frequency of shopping at supercenters and supermarkets was associated with the healthfulness of foods purchased but, similar to our results, was not associated with BMI [18
]. In a study situated in two mostly African-American neighborhoods with low-income residents in Pittsburgh, Pennsylvania, Dubowitz et al. (2015) did not find a relationship between shopping frequency and BMI or dietary intake quality [15
An unexpected finding in our study was that using a supercenter (e.g., Walmart and Kmart) or a warehouse club (e.g., Sam’s Club) as the primary food store was associated with significantly higher BMI levels compared to shopping at a supermarket. Of note, 27% of study participants shopped at supercenters or warehouse clubs for store 1, 22% shopped at these stores for store 2, and 14% used them for store 3, with the vast majority of participants shopping at supercenters (i.e., only 1–3% shopped at warehouse clubs). Interestingly, Jilcotts-Pitts et al. (2012) and Minaker et al. (2016) did not see this distinction [18
], and the initial positive association of supercenter use and BMI observed by Dubowitz et al. (2015) was attenuated after adjustment [15
]. Drenowski et al. (2012) reported significantly higher rates of obesity among shoppers at medium- or low-price supermarkets than at higher-price supermarkets in a study in Seattle. Low-price supermarkets predominantly included FredMeyer and Albertson’s, and higher-price supermarkets included Puget Consumer Co-op and Whole Foods, among others [12
]. Ghosh-Dastidar et al. found lower store prices in the utilized stores (but not distance to these stores) to be associated with higher obesity in the aforementioned Pittsburgh study [26
]. In a study of residents of the Paris, France, region, Chaix et al. (2012) reported that shoppers at hard discounters (e.g., Aldi) with middle to lower levels of education had significantly higher BMI compared to those who did not shop at hard discounters [11
]. Other studies of supercenter stores and BMI are more indirect: Gustafson et al. (2011) showed that individuals residing in a census tract containing a supercenter were heavier than individuals in a tract without a supercenter [40
]. All of these findings raise challenging questions as to the inferences that can be made, which we discuss more below.
Additionally, we found that using a smaller store, i.e., a store in the “other” category (for example, a convenience store, drugstore, dollar store, small grocery store, or specialty store) as a third store was associated with significantly lower BMI compared to shopping at a supermarket. Minaker et al. (2016) comprehensively describe reasons for different store type choices in a study in Ontario, Canada [21
]. They found that proximity and convenient hours were the most frequent reasons for choosing a convenience store, whereas quality and availability of specific foods were the key reasons for specialty store selections. Although they found no association of convenience store utilization and BMI, they found a significant inverse association of specialty store use and BMI. Given that we did not assess reasons for store choice, we can only hypothesize that choosing a smaller store as one’s third store may reflect a shopping behavior pattern that tailors selection of store type to the purpose of shopping trips and that the ability to be this selective may also indicate easier access to transportation.
A further interesting finding emerged from studying the association of a resident’s perceptions of his/her food environment with BMI, for which we utilized three different scores, one a composite score of the availability of healthful options in the neighborhood, one on availability of fast food outlets, and a more general question on whether access to food shopping was a problem. We found that the less a person considered lack of access a problem, the lower their BMI, or, conversely, the more a person considered lack of access a problem, the higher their BMI, and these findings were independent of the other food shopping behaviors and food environment characteristics, as well as the other two perception measures. The direction of these findings is consistent with both longitudinal and cross-sectional findings from MESA [41
], which reported that residents that perceived their neighborhood as the worst in terms of healthy food access had a significantly higher obesity incidence and prevalence [41
]. One potential interpretation of this finding is that participants perceiving lack of access as a significant problem are expressing barriers constraining their food shopping-related choices, such as the foods purchased, frequency of shopping trips, and transportation to stores. We have previously shown that residents’ perceptions of relatively lower availability of healthful foods and relatively greater lack of access are not particularly correlated with spatial availability measures of the built food environment but are strongly associated with household socioeconomic characteristics, including food insecurity [43
However, arriving at a causal interpretation from observational research on food shopping characteristics is complicated by the fact that it is entirely possible that people who shop at a particular store type (e.g., supercenter or discount store) may differ systematically from those who shop at different types of stores, including in their individual socioeconomic characteristics but also in attributes of the economic, social, and cultural system in which they reside [44
], which in turn could influence the types of foods purchased and eaten, leading to obesity [45
]. This phenomenon can manifest as selection bias (e.g., differing sampling fractions of study participants who shop at a certain store type and eat healthful foods). Conversely, it could manifest as an endogeneity problem, either through uncontrolled (i.e., residual) confounding (also known as omitted variable bias) or through reverse causality (i.e., persons with a higher BMI being more likely to choose to shop at supercenters than supermarkets).
Using a unique study design as a selection of shoppers nested within a selection of stores, Chaix et al. (2012) were able to determine that shoppers within the same store type were significantly more similar in terms of socioeconomic characteristics and BMI than shoppers between different store types [11
]. Courtemanche and Carden (2011) conducted a contextual study relating the county-level prevalence of Walmart supercenters with individual-level BMI and obesity rates and found after adjusting for endogeneity bias (which they interpret as arising from the non-random nature of Walmart’s locations) using an instrumental variable approach [46
] that one additional supercenter per 100,000 residents was associated with an average 0.237 kg/m2
higher BMI, which, given their population’s characteristics, equated to half a pound [47
]. It is conceivable that given sufficient incentives, individuals could be randomized to utilizing certain grocery stores and followed up over time to observe potential changes in BMI; however, unless adherence to stores is excellent, such a randomized trial could be equally subject to bias.
Several limitations and strengths are worth mentioning. The data presented here are baseline data from a study designed to evaluate a natural experiment. Thus, the study’s size is a function of the recruitment goals, which aimed for a certain sample size and power to detect change over time (and achieved 94% of the enrollment goal of 560 at baseline). Thus, we cannot rule out the possibility that the participants of this study are not representative of all residents of the recruitment area. We can say with certainty, however, that the participants represent a very low-income and largely minority population. This study did not collect data on food prices at all utilized stores, only those at a small subsample of the five most utilized stores, and so we cannot reach any conclusions with respect to an independent role of food prices at the store level and their relationship with BMI. Furthermore, this study lacks data on the types of foods purchased by participants, which is information in an important mediating pathway. Our study focused on BMI as an outcome, instead of overweight/obesity, because 80% of our population was overweight or obese. The key distinguishing feature of our study is that we assessed characteristics of the three most-frequented stores rather than just the single most-frequented store and additionally considered food acquisition at non-retail locations such as food banks/pantries and churches/social service organizations. We also had data available on transportation to the primary store, although we did not find a significant association between this variable and BMI.
In summary, this study evaluated the association of three distinct sets of food environment-related measures—food shopping and acquisition characteristics, perceptions of the food environment, and GIS-based built environment measures—with BMI levels. Shopping at a supercenter as one’s primary store and reporting that one considered lack of access a significant problem were both associated with higher BMI levels, whereas selecting a smaller store as one’s third most-frequented grocery store was associated with lower BMI levels. We have previously shown in this population that 76% of residents shopped at supermarkets or supercenters for both store 1 and 2 [27
The positive finding on the supercenter-BMI association raises the questions of whether these types of stores do not have as good or as large a selection of healthy food as supermarkets, if healthier food options within supercenters are placed at a price disadvantage compared to less healthy versions of the same type of food, or if the customers utilizing the stores predominantly make less healthy purchasing decisions. Several studies have shown that supercenters frequently offer among the lowest grocery prices in a given region and thus disproportionately attract customers who are very price conscious [48
]. Moreover, supercenter utilization was associated with lower educational attainment, lower income, and greater household size in a US telephone survey [45
]. In addition, several studies have shown that supermarket utilization was associated with higher income, including in a sample of African-American women from a low-income neighborhood without supermarkets in Detroit [16
]. A partial answer to these questions challenging causality could be provided by a study that assesses the nature of purchases and controls for them while evaluating the same research questions we posed here. However, this still leaves open the possibility of residual confounding through foods purchased at other types of food outlets and dietary choices made by individual members of a household and other unmeasured lifestyle factors.
The implications of our study are multi-faceted: First, these findings challenge the concept [50
] that limited spatial access to healthy foods, i.e., greater distance to large grocery stores, is an important cause of the obesity epidemic or, conversely, that eliminating food deserts will help curb the obesity epidemic. The vast majority of participants in our study shopped outside of their residential census tract. Food desert residents clearly experience increased grocery shopping burdens, including on average 7.5 extra miles traveled for shopping per week, plus higher fuel and time expenses, as shown in one of our previous studies in South Carolina [10
]. However, the present study does not offer evidence that distance is associated with obesity in the population studied here, even though 55% do not travel to their primary store with a car of their own. Second, this study’s findings call into question that all large grocery stores, including supermarkets and supercenters, are beneficial in addressing the food desert and obesity problem. Our study suggests that the food-purchasing behaviors in different types of large grocery store (e.g., supermarkets, supercenters, and warehouse clubs) and motivations and constraints in store selections should be studied in more detail, particularly in under-resourced populations. If future studies find that supercenter customers tend to purchase less nutritious foods, creative new approaches would be needed that combined tailored interventions with system-level incentives toward healthy behaviors, because individually targeted point-of-sales nutrition education programs have not been particularly successful [51