Improving spatial access to healthy food retailers has emerged as a novel approach in public policy in the United States, complementing longstanding policies on improving economic food access. Various agencies have developed measures of community food access for purposes ranging from surveillance to policy implementation. The U.S. Department of Agriculture, Economic Research Service (USDA-ERS) has issued a definition for food deserts, which is embedded in the online ERS Food Desert Locator and linked to the Obama Administration’s Healthy Food Financing Initiative. The Centers for Disease Control and Prevention (CDC) have reported on U.S. Census tracts with healthier food retailers in the context of their State Indicator Report on Fruits and Vegetables, and The Reinvestment Fund (TRF), a Philadelphia, PA-based community-investment group, has defined limited supermarket access areas.
A brief comparison of the published methodologies of these measures of community food access reveals that they rely on very different, secondary data sources listing retail food outlets, and are based on unique sets of criteria and methods. These measures of community food access have not been compared systematically in terms of the areas identified or the size of the populations that are designated a living in areas with poor food retail access. Furthermore, previous research by the authors and others has demonstrated that secondary data sources for food retailers contain substantial amounts of error including undercounts, overcounts, and geospatial inaccuracies. Errors in secondary data may bias measures of community food access, but to date, very little is known regarding the magnitude of such influences.
This study replicated the USDA-ERS food desert measure, CDC non-healthier retail tract measure, and TRF limited supermarket access area measure in order to address the following research questions:
(1) How do these three measures of community food access compare with respect to the areas identified as having poor food access and the populations affected, if a single, validated data source of the food environment were used?
(2) To what extent do inaccuracies in secondary data sources influence these three measures of community food access?
The replication algorithms were programmed in ArcGIS 10.0 and related extension software based on the criteria and definitions for each measure contained in government or agency publications. First, the replication was conducted using a verified (ground-truthed) database of food outlets in an eight-county region of South Carolina, covering approximately 5,575 square miles and a population of more than 620,000. Each Census tract received a designation with respect to being a food desert, a non-healthier retail tract and a limited supermarket access area. The replication based on verified served to address the first research question and also served as the reference for the second research question. Subsequently, the replication was conducted using two commonly used secondary data sources listing food outlets (Dun & Bradstreet, InfoUSA) and accuracy measures were computed.
Comparison of the three measures of community food access with respect to the areas identified revealed that according to the USDA-ERS criteria, only 10 percent (n=15) of 150 Census tracts in the study area were designated as food deserts, compared to 28.7 percent (n=43) designated as non-healthier retail tracts by CDC, and 29.3 percent (n=44) limited supermarket access areas by TRF. The comparatively smaller number of USDA-ERS food deserts was due to the fact that these areas were really low-income areas that additionally suffer from low access to healthier food outlets, whereas the CDC and TRF measures focus mostly on low access. Despite these differences, the geographic overlap between the three measures was quite high (USDA-ERS versus CDC: 71 percent of tracts; CDC vs. TRF: 77 percent; USDA-ERS and TRF: 65 percent). The population estimated to be residing in areas with poor access to healthy food choices ranged from 50,085 to more than 201,300.
Furthermore, this study found that inaccuracies in secondary data sources used by agencies influence the designations as areas with poor food access. With respect to areas designated by the USDA-ERS as food deserts, relying on the secondary data sources identified fewer food deserts than actually existed in the 8-county study area. However, there was little over-ascertainment compared to the reference data. With respect to the CDC, although the secondary data sources identified nearly the same total number of tracts identified as non-healthier retail tracts by CDC as the reference data, the data sources did not consistently identify the same tracts, due largely to under-ascertainment and some over-ascertainment of areas with poor food access. Secondary data sources showed little under- and over-ascertainment for the TRF measure.
In summary, this study revealed marked differences between three measures of community food access in terms of their intent, the underlying methodologies, and in the resulting designations of areas as having poor food access. This suggests a need for clearer communication of conceptual differences and potential harmonization efforts. Furthermore, this study demonstrated that inaccuracies in the secondary data sources used by various agencies to create their respective measure of community food access influences the results. In general, under-ascertainment of areas with poor food access seemed to be a somewhat more pronounced problem, especially for the USDA-ERS food desert measure, rather than over-ascertainment. This finding suggests that prior to initiating efforts to improve the food retail environment in a given area, stakeholders would be well advised to verify that the targeted area in fact has as poor food access as the community food access measure may suggest.