Developing a Food Desert Index Using Spatial Analysis
The GIS component of this study accomplishes two tasks that serve different purposes. First, an integrative GIS analysis (“Food Desert Analysis”) using advanced quantitative techniques is conducted to furnish a multi-dimensional understanding of the food accessibility issue, in terms of food deserts in the study counties. The second task is to develop a qualitative mapping tool (“Food Desert Locator”) in the form of a geo-processing model in ArcGIS that enables the user to perform quick and less complex examination of the food access landscape in the study area. Due to its emphasis on a comprehensive assessment, the Food Desert Analysis must be supported by large quantities of data, more sophisticated analytic techniques, and operator supervision throughout the process. Therefore, the client and other users of the findings of this study might have difficulties replicating the food desert analysis. Therefore, the Food Desert Locator is provided as a user-friendly substitute to apply in resource-limited or efficiency-driven endeavors.
The Food Desert Analysis requires ESRI’s ArcGIS 10.2 for Desktop software with Advanced License, Geostatistical Analyst extension, Network Analyst extension, and Spatial Analyst extension. The Food Desert Locator requires the same software and licensing with only Network Analyst extension.
Food Desert Analysis Methods
Designed to be an overall assessment of the study counties’ food accessibility condition, four types of non-geographic barriers—Geographic, Economic, Vulnerability, and Cultural and Informational barriers—along with geographic barriers, are utilized to evaluation food accessibility of places. In particular, following the works of Parsons (2012), a total of six variables of socioeconomic characteristics are identified and grouped into three non-geographic barrier groups (Table 1). Quality and resolution of the input data vary depending on US Census Bureau’s data publication (Table 1). In order to unify input format and resolution, non-geographic variables are converted to raster layers using ArcGIS’s Geostatistical Analyst extension. Meanwhile, two variables are used to capture geographic barriers—street distance to the nearest food retailer, and household availability of vehicles. The former is calculated using ArcGIS’s Network Analyst extension, output in polygon vector format. The latter is also converted to raster using the same method as with non-geographic variables.
The final products of the Food Desert Analysis consist of four summary maps, three showing the spatial distribution of each non-geographic barrier score, and a fourth one summarizing these three into a “final score” map. Overlaid with distance-based food desert polygons (result from geographic barriers), the final score layer represents the effects of all three non-geographic barrier combined, and provides a comparable view of these results along with geographic barriers.
Non-Geographic Barrier Scores
The resulting Economic, Vulnerability, and Cultural/Informational Barrier Scores (“barrier scores”), along with their arithmetic average, the Final Barrier Score (“final score”), are summarized in Table 2 and Figure 1. As discussed in Methods, all scores have a possible range of 1-100. The actually calculated barrier scores have basically realized their full ranges, with only one insignificant exception (Table 2). The final score, however, has a realized range of 8-89, suggesting that places having either three lowest scores or three highest scores at the same time are not found in the study counties. In other words, even places experiencing extreme stresses of two barriers are somewhat moderately stressed with the third barrier.
Given the range of the barrier scores, the means smaller than 50 imply positively skewed distributions (Table 2, Figure 1). Among the three barrier scores, the distribution of the economic score is more symmetric, having a more pronounced center peak and lower frequencies at the extremes. This shape of distribution is typical among economic characteristics in America. The vulnerability and cultural/informational barriers scores have higher low-score peaks, indicating large areas free from these two types of barriers. However, their greater standard deviations, lack of center peaks, and the plateau tails at the high extremes interpret into relatively large areas inhabited by vulnerable populations (elderly, children, disabled), under-represented cultural groups (African-American population), and population of limited informational access to healthy food choice (population of limited education attainment). In contrast, areas of medium stress are not dominant for these two barriers, as opposed to a symmetric distribution with a large center body and small tails. The vulnerability and cultural/informational barriers are primary contributors to the slightly skewed shape of the final score distribution (Figure 1).
Figure 1: Distribution of Barrier Types
Spatially, non-geographic access barriers are commonly found around urban areas of Plymouth in the north and Washington in the south. Stressed areas are also found in the lower Beaufort County south to the Pamlico River, and northeastern Beaufort County near the Hyde County border (Figure 2). Worth noting is the spatial similarity of economic barriers and cultural/Informational barriers observed in Figure 2, while the vulnerability barriers share less distribution patterns in common with the other two. The final score map (Figure 2) is a spatial aggregation of the barrier score maps.
Figure 2: Non-geographic Barrier Score Maps
The geographic barriers, in form of distance-based food deserts, and non-geographic barriers, in form of final score surface, are overlaid on the same map (Figure 3) to facilitate easy comparison of these results. As is shown, “hotspots” stressed by non-geographic barriers (low final scores) tend to occur in urban and densely populated areas. These are also the places where food retailers tend to be found. As a result, areas suffering both geographic barriers (difficult to reach a food retailer) and non-geographic barriers (socio-economically stressed) are largely reduced. The areas that experience both physical barrier to food retailers and stresses of non-geographic barriers (termed “all criteria food deserts” or “food deserts”) are found in three clusters (Figure 3), totaling approx. 34.3 square miles. The largest cluster is located on the southern bank of Pamlico River in Beaufort County. A smaller piece of food desert area is near the Hyde County border in NE Beaufort County. The only food desert identified in Washington County is in urban Plymouth. Notably, the other non-geographically stressed urban center, i.e. Washington in Beaufort County, is not identified as a food desert due to the clustering of grocery stores near the city.
Figure 3: Geographic Barrier Map versus Non-geographic Barriers Map