The goal of this article was to describe the relationship between the prices charged by stores for healthy food items and the characteristics of the surrounding communities. We found that holding racial composition constant, stores located in neighborhoods with higher income residents charge more for fresh fruits and vegetables. This is consistent with previous results in other urban settings [24, 25]. For fruit, much of the difference in cost between stores was accounted for by differences in the price of one item, avocado. For vegetables, however, the relationship between neighborhood income and price was present for many of the most common types.
A positive relationship between the affluence of surrounding communities and the cost of fresh produce (holding racial composition constant) can be explained by a number of forces. First, fresh produce is a normal good, i.e. households purchase more fresh fruits and vegetables as their incomes increase. Some of this increase can be attributed to a general increase in all purchases as income goes up. It is also possible that households with higher incomes shift their consumption of fruits and vegetables toward fresh items, instead of canned or frozen versions. Of course, these explanations are not mutually exclusive and could be occurring concurrently. Regardless of the mechanism through which income causes demand for fresh produce to increase, basic economic theory predicts that as demand for fresh produce increases, the price of produce will also increase.
A second explanation is that fresh produce is not a homogenous product. Quality is also a normal attribute so that as incomes increase, individuals attempt to replace low-quality items with higher quality version. Hence, differences in prices between stores may reflect differences in the freshness of items.
Finally, stores in rural areas may face less price competition. In densely populated areas, higher demand for fresh produce may not lead to appreciably higher prices since there are many stores competing for the same set of potential consumers. Since rural areas typically exhibit low population densities combined with large distances between population centers, the number of stores that are reasonably accessible to residents will be smaller, reducing the amount of competition.
Holding the median household income constant, stores located in neighborhoods with higher proportions of African American residents also tended to charge a higher price for fresh produce items. Part of the explanation may be that greater proportions of African Americans reside in CBG where there is greater access to small grocery and convenience stores, where prices are higher than charged in the larger supermarkets and supercenters [16, 17, 32]. In contrast, the proportion of Hispanic residents did not exhibit a statistically significant relationship to affordability. As is well-documented, African Americans in the United States suffer from higher rates of nutrition-related illness such as diabetes and are much more likely than Whites to be obese [37, 38]. This is also evident in the Brazos Valley region, where one study found that the obesity rate for Blacks is 46.4% compared to 32.0% for Whites . The affordability of fresh produce may be one factor that contributes to the disparity.
Univariate analysis revealed that failing to control for income and racial composition simultaneously can influence coefficient estimates, and thus policy implications. For example, our multivariate regression results suggest that a policy that sought to reduce income inequality between neighborhoods with high and low proportions of Black residents could drive up the price of fresh produce, so that in real (purchasing power) terms, neighborhoods with higher proportions of Black residents would remain relatively disadvantaged. This clearly important effect is missed in the univariate analysis.
In addition to providing a fuller description of affordability differences by economic status and racial/ethnic composition in rural areas, the current paper also makes several methodological improvements over previous work. First, our data is from a census of food outlets that utilized ground-truthing. While ground-truthing methods have been used elsewhere, they often are based on just a sample of stores [23, 31], cover a relatively small geographic area [21, 27] or are employed in areas with limited economic or racial/ethnic heterogeneity [26, 40, 41]. In our application, all stores in a large rural region that is economically and demographically diverse are present in the dataset.
Second, the analysis confronts the common problem of missing prices using a price imputation strategy that is more firmly grounded in economic and statistical theory than has previously been employed, addressing a potential source of bias [21, 34]. Although results with our regression imputation strategy were similar to those using mean imputation in this particular analysis, in other contexts ignoring the profit maximizing behavior of store-owners may not be benign. More generally, researchers in this literature should take greater care with their imputation decisions by explicitly recognizing the underlying assumptions inherent in any method and checking the robustness of their results.
The analysis suffers from several limitations. First, the number of supermarkets and grocery stores in the rural Brazos Valley region is relatively small and future work should consider canvassing a larger area to increase the precision of estimates. Doing so could not only increase the number of observations in an analysis similar to the one undertaken here, but also allow for separate regressions for urban and rural areas. There are obvious trade-offs between completeness and breadth of coverage and small sample sizes are shared with previous work that employed ground-truthing [16, 20, 40, 41]. While suggesting that larger data collection efforts be undertaken, we also acknowledge that the cost associated with completing a census of food stores should not be underestimated
Second, we are unable to translate differences in local affordability into differences in purchasing behavior. If transportation costs are low, then local affordability may not influence actual purchasing or consumption decisions. This could be the case if individuals lived in rural areas, but commuted to an urban center for work. The six counties in the rural BV region are circumjacent to Brazos County, which has a population of almost 150,000 concentrated in the cities of College Station and Bryan and home to Texas A&M University. In future work, we wish to consider how the affordability of neighborhood food prices influences the decision of where to shop, and the activity/travel patterns of rural residents.
A closely related issue is translating differences in affordability into differences in eating behavior. If higher prices do not affect consumption patterns, then attempts to lower the cost of fresh produce may lead to overall gains in welfare, but will not influence nutrition-related health outcomes. This may be particularly salient in explaining racial and ethnic disparities, as the association between neighborhood socio-economic status and consumption of fruits and vegetables is stronger for Blacks than Whites . Again, this is an area of future research.
Third, it must be acknowledged that fresh whole items from food stores are not the only source of fruit and vegetables, though as stated previously, the majority of fruit and vegetable consumption is in the form of fresh whole items. In food stores, fruit and vegetables can also be purchased in frozen, canned, dried and juiced forms. Additionally, fruit and vegetable consumption may also occur in restaurant settings. Since the nutritional value of consumption likely varies by the form consumed, the affordability of these different options, both in absolute terms and relative to each other, is also worthy of future study.
Fourth, sales shares were not available for the stores in our sample, and thus we were unable to weight store-level observations in the regression analysis. Future data collection effects should do so in order to account for differences in the relative importance that each store plays in the actual purchasing decisions of households. Alternative methods include weighting stores by a combination of cash registers in service and hours of operation , but this information was not collected during surveying.
Finally, while our imputation strategy allows us to calculate a hypothetical measure of affordability for stores that do not sell all items, these stores may typically exhibit limited availability. We are unable to document a possible variation in produce prices due to seasonal variation. Researchers and policy makers should keep both aspects--availability and affordability--in mind when considering improvements in the food environment.