An online alternative to measuring food environments

Introduction: Inaccuracies in secondary data sources on food environments compared to field censuses have been demonstrated many times. Despite the policy implications of erroneous data, field censuses are often too costly and time-consuming to be practical. We proposed using online storefront images to categorize food retail establishments as a more feasible alternative to accurately measure food environments, and aimed to assess geographical variability in errors.

Methods: Using online storefront images, supermarket websites, and local knowledge in conjunction with a database of food retail establishments, we identified and categorized food stores in New York City neighborhoods with the lowest reported levels of fruit and vegetable consumption. We compared our classification to a list based on the U.S. Department of Agriculture (UDSA) food desert measure (using solely secondary data sources). We determined spatial heterogeneity in the error using spatial statistical methods.

Results:  Compared to USDA criteria, our method identified more supermarkets (466 vs 55), grocery stores (302 vs 239), and convenience stores (1570 vs 1097), and fewer fruit/vegetable specialty stores (225 vs 292). The secondary source overcounted the presence of food stores by 13% (found to be duplicates, non-food stores, wholesalers, or invalid/residential addresses). We anticipate that data quality will vary significantly by neighborhood due to the level of miscounting and the concentration of particular categories (e.g. wholesalers) in particular areas.

Discussion: Secondary data sources provide variably inaccurate information about food environments. Availing of online maps and images to identify food stores may become a cost-effective approach for researchers and policymakers.