INTRODUCTION

Access to food in the United States seems to be universal, but access to fresh foods in many places in the U.S. is incredibly limited and often a large distance from many residences. I wanted to visualize these “food deserts” to compare them to other demographics based on location in order to see what  correlation exists. Mapping demographics like race or household size in comparison will show which demographics often overlap geographically with food deserts in New York State.

• What visual relationship will appear when mapping New York State food deserts in conjunction with 2010 census demographics?

INSPIRATION

This first visualization is the map made by the USDA to display the data they collected about food deserts. It is a very basic visualization that shows a minimum amount of information. It solely uses the data from the food desert dataset as that is the sole narrative they are telling. I want to use this data in conjunction with other data to see if a comparison will shed light on any potential relationships between food deserts and other factors.

The second map is from scientific america and combines the food desert data with data on household vehicle ownership, obesity, and diabetes. While it shows four factors displayed on a map of the U.S. by census tract, it does not overlay these maps. Since they are separate it is difficult to see exactly where the different variables overlap. I want to layer my data in hopes to make an easier visual comparison.

The final visualization is similar to the previous map, except it focuses on a single metropolitan area, Chicago. Similar to my goals, it shows a smaller geographical area which makes it easier for the user to digest this information and do sight comparisons. I can begin to see some obviously overlaps between to the two maps, it still does not layer the maps directly onto each other.

METHODS

Data

The first source for data came from the United States Department of Agriculture’s Economic Research Service. This data divides NYS by 2010 census tracts with several variables for each tract relating to food access, including by age, distance from supermarket, vehicle access, and income. Most of these variables simply give the number of individuals or households in each tract that meet the criteria named. The second dataset came directly from Census.gov. Census data from 2010 includes information such as age, race,  or gender. This includes values as total number of people per tract and as percent of the whole tract. With over 350 columns, I eliminated the variables that I was not interested in mapping before proceeding. The 2010 census data included the shapefiles to build the map, which will also match up with the USDA food desert data.

CartoDB

To begin building a map, I first uploaded the shapefiles of the 2010 census tracts. Since both the census demographic data and the food desert data include census tract Ids, I was able to merge them on CartoDB before creating any maps. I started by adding a choropleth layer for food deserts, which colors census tracts on a grey scale gradient based on how many people in each tract are considered living in a food desert; both distances of 1 mile or 10 miles away from a food supply were tested for possible results.

I tried adding many different demographic choropleth layers to the map to see which variable made the most compelling visuals. These demographic variables included tract population by race (Black/African American, White, & Hispanic), residence (renter or owner), and vehicle access. Using a light red color gradient on this layer meant that when the demographic was high in the same tract as a food desert, the colors would multiply and produce a darker red. This was the best case result for mapping these two data sets, producing a visualization that uses color effectively without losing much of the original meaning in the data.

RESULTS

After mapping several combinations of variables, I had moderately successful visualizations as a result. While no one visualization completely satisfied my original research question, there was some interesting correlations. One example occurs in central Buffalo. There is an area that very closely overlaps with both a food desert and a majority African American population which causes the red & blue choropleth layers to overlay into purple. This areas was a successful attempt to map the correlation of these variables in a way that clearly illustrates a relationship. 

Overall the map that layers households with low vehicle access and food deserts at a 10 mile distance produced the most complete picture. It’s easy to see the areas where these two conditions overlap, but does not incite any revelations about the data. I chose to include racial demographics in each census tract popup, so that the specifics of each tract are available for consideration even if they are not part of the final visible layers.

FUTURE DIRECTION

An important future direction for this data would be running statistical analysis on both datasets to determine correlation between any variables.  Tests could if any of these demographics, or groups of variables, have any statistical relationship to living in a food desert. A Canonical Correlation Analysis could break down this multi-varied relationship and create synthetic multidimensional variables that will give a more nuanced interpretation of the data. Mapping this data after running statistical tests could give a better idea of which variables correlate in a way that would make an informative visualization.