Client Meeting October 26th

Food Security: “Access by all people at all times to enough food for an active and healthy life” - World Bank, 1986

Purpose Statement

This storymap will examine the GIS processes and replicability of the method for identifying food deserts presented in the article, “Towards the development of a GIS method for identifying rural food deserts: Geographic access in Vermont, USA” by Jesse McEntee and Julian Agyeman.

Key Concepts and Disciplines (Tags)

Article Summary

“Towards the development of a GIS method for identifying rural food deserts: Geographic access in Vermont, USA” by Jesse McEntee and Julian Agyeman was published in January 2010 in Applied Geography. The article states that while there is agreement as to the concept of a food desert, there is no systematic method for identifying them. Thus, the paper seeks to identify a GIS approach to identify food deserts. This approach involves manipulating spatial data to provide quantitative analysis. In order to develop and test their method, which averages the distance of residential units to grocery stores to identify distances to accessible food, they focus on the rural state of Vermont. 

The authors argue that an easily applicable approach for identifying food deserts is essential to public health research, as opposed to previously costly qualitative food index surveys. Such a strategy would allow researchers to address the three types of food access: informational access (in which educational, cultural, and social constraints influence food choices), economic access (in which financial elements impact food acquisition), and geographic access (in which distance and geographic features impact food access). As a GIS method, the paper’s methods will largely be beneficial to those attempting to address geographic accessibility.

Data Collection

To find the data for this project, we followed the methods expressed in the original article. McEntee and Agyeman listed the food retailers as “North American Industry Classification System number 44511 - Supermarket and Other Grocery [except Convenience] Stores larger than 2500 square feet.” They used  Reference USA  to access this data, which was easily accessible and replicable. However, the site would only let us download grocery stores 250 at a time, so we needed to use the pandas package in python in order to coalesce the two csvs into one file, with 294 grocery stores. This step indicated that we came up with more grocery stores since the article was published in 2010, when only 142 grocery stores were collected. 

To find housing data, we used their method of the  Vermont Center for Geographic Information , and found 258,711 compared to their number of 231,894 residential locations in 2010.

Finally, in order to track the ability of people to get to a grocery store, we also needed to find road data and census tracks. Unfortunately, their description of finding road data was poorly defined, so instead we turned to this  Esri dataset  which holds centerlines for roads in Vermont. We then also used the Census MAF/TIGER database to define the census tracks.

Methods

We began the geoprocesses outlined in the paper by loading each dataset into ArcPro. We first needed to limit the housing dataset by the following attributes using the “Select Layer by Attributes Tool” in order to ensure the captured building units only contained residential units: single family dwelling, seasonal home, residential farm, other residential, nursing home/ long term care, multi-family dwelling, mobile home, condominium, commercial with residence. Unfortunately, the paper was not entirely clear on whether it counted certain types of parcel units, such as lodging/B&B/Hotel/Motel/Inn, or Institutional Residence/Dorms/Barracks, as part of their dataset. We chose not to include these units. We then exported the selection as a new layer (layer from selection tool), titled vt_houses.

Next, we projected each of the existing layers and the basemap into NAD_1983_StatePlane_Vermont_FIPS_4400. This step was not discussed by the authors, but seemed necessary in order to ensure accuracy once we began using geoprocessing tools. 

The authors state that their next step was to use the “Network Analyst Extension Closest Facility” tool in ArcMap 9.1. Unfortunately, ArcMap is no longer available. A similar tool, Find Closest Facilities, is available in ArcPro, but had a limit of 5,000 data points, which was not feasible for the magnitude of the data we collected. As a result, we decided to subset the data into groups for analysis.

We began by selecting two census tracks, numbers 9462 and number 9635. These census tracts were selected for primary analysis, the original paper designated one of them as a food desert and the other not, they neighbored each other, they do not border another state, and they each had less than 1,000 housing units for processing speed. As a result it is necessary to recognize that future analysis will likely take longer, and has been factored into our project timeline.

We then used the "Find Closest Facilities" tool, designating the Vermont Grocery Layer as facilities, and the incidents as the housing units. This tool was run once for each of the subsetted census tracks. The mode selected was driving distance in miles, and everything else was left as default.

After allowing the tool to run, a routes layer was created, routing each house to the closest grocery store. Then we used the summary statistics tool to calculate the mean total miles of each house to its closest grocery store. For census tract 9642, the average distance from residence to grocery store was 2.27 miles. For census tract 9635, the average distance from residence was 6.28 miles.

According to the paper by McEntee and Agyeman, an average distance of over 10 miles from residences to grocery stores is the minimum qualification as a food desert. As a result, neither of these tracts, according to our calculations, would count as a food desert.

On these two census tracts alone, our results did not match those of McEntee and Ageyman. There are a few potential causes of this discrepancy.

• First, we were unable to use the same exact tool as the original paper, as their process involved ArcMap.
• Second, the data we used was updated to current times, so new grocery stores and houses could have affected general outcomes.
• Third, we suspect that the 10 mi designation of a food desert may need to be updated based on current standards for food deserts.

Timeline of Accomplishments

• September 30th- October 11th: Researched, found, and formatted required data elements
• October 12th - October 20th: Analysis and testing on first two census tracts

Estimated Timeline of Future Accomplishments

• It is estimated that completing analysis on the remaining census tracts will take one additional week.

Conclusion of Overall Replicability

Overall, their process was replicable and clear, despite our differing results. Analysis on the remaining census tracts may reveal more matches between our results and those of the original paper.

Their parsimonious analysis technique allowed for us to easily follow their methods, although their results were not replicable since they did not publish their original datasets and we were thus forced to use updated data. The authors were not responsive to requests for the data they used.