Geographic Boundary Definitions and the Robustness of Common Food Retail Environment Measures

Understanding the relationship between the retail food environment and individual and community health is important for addressing questions about diet-related disease and food security. A wide range of methodological approaches have been used to understand this complex relationship, but the studies have resulted in a lack of consensus on how and whether the food retail environment influences health outcomes and food choice. Inconsistency in measuring the food environment could explain inconsistent results. We calculate five different boundaries of the food retail environment to investigate the robustness of boundary measures. These five boundaries include an administrative boundary (census tract of residence), two Euclidean boundaries, and two network area boundaries. We compare four different aspects of the food retail environment—availability, accessibility, affordability, and realized purchase behavior—across these five fundamentally different geographic boundary definitions. Using the National Food Acquisition and Purchase Survey, Nielsen TDLinx, and Information Resources, Inc. retail and consumer data, we find that food environment measures using administrative boundary definitions are statistically different from all other boundary definitions. We find general similarities within the two Euclidean distance and two network area boundary definitions. Only a small portion of food-at-home purchase events and expenditures by households occur within their own census tract boundary, suggesting the boundary definition has limitations and households are mobile in their food acquisitions.