Spatial Planning to Examine Regional Racial and Ethnic Disparities in Obesity and Diabetes by Analyzing the Supplemental Nutrition Assistance Program (SNAP) Retail Environment: Applying a Machine Learning Approach
Date of Award
Doctor of Philosophy (PhD)
Planning, Design, and the Built Environment
James H. Spencer
In the United States, food insecurity has been an ongoing concern nationwide, with over 17.4 million Americans affected. The United States Department of Agriculture (USDA) Supplemental Nutrition Assistance Program (SNAP), commonly known as “food stamps,” aims to improve food security by providing financial assistance on food purchases to income-eligible households. SNAP participants redeem their benefits at SNAP-authorized retailers, including supermarkets, small discount stores, convenience stores, corner stores, and pharmacies. The retail food environment influences food purchasing behavior, thus contributing to poor diet. In the United States, low-income residents, including SNAP participants and minority residents, are more likely to have diet-related diseases, including obesity and Type 2 diabetes, due to the consumption of inexpensive energy-dense foods. However, there have been no studies to date that examine the relationship between the SNAP retail environment and low-income, minority residents. Therefore, this study has three aims:
Aim 1: Examine obesity and diabetes prevalence and their relationship to urbanicity and SNAP retail food environment.Aim 2: Analyze spatial mismatch between healthy SNAP Retailers and high SNAP participation communities. Aim 3: Identify regional differences in the SNAP retail environment and its relationship to health, race/ethnicity, and SNAP participation.
SNAP-authorized retailer information was downloaded from the United States Department of Agriculture, Food and Nutrition Services SNAP Food Retailer Locator website. The list was based on the 2017 federal fiscal year and included store name, address (street, city, county, and ZIP), and latitude and longitude coordinates. Each store was categorized by retail type based on the North American Industry Classification Systems codes. Small grocers were defined as grocery stores with less than three employees, which were identified through the Business Analyst tool from ESRI ArcGIS. Each SNAP-authorized retailer was coded as healthy or unhealthy based on the Centers for Disease Control and Prevention criteria.Census tracts were identified based on the Centers for Disease Control and Prevention (CDC) 500 Cities project, which aimed to compile epidemiologic data of chronic disease risk factors, clinical preventive services, and health outcomes at the census tract level for the 500 largest cities in the United States. This study included at least one city in each of the 48 contiguous states and the District of Columbia. Nearly 500 cities, including 22,729 eligible census tracts, were used for this study. Each census tract from the 500 Cities Project includes a measure for diabetes prevalence and obesity rate. Race/ethnicity, SNAP participation, and other socioeconomic were collected from the American Community Survey using 2017 5-year estimates data at the census tract level. The analysis examined predictive relationships of poor health, racial/ethnic disparities, and access to SNAP-authorized retailers. Each construct included multiple variables for the measure, thus increasing the robustness of the study design. Multivariate regression analysis was performed applying a forward stepwise approach. However, given the number of predictors and observations, machine learning techniques allowed to quantify uncertainty, identify best predictive models, and test model performance in a more efficient way. Therefore, while forward stepwise regression was used to build a full model, residual diagnostics and machine learning techniques were applied to identify best predictive models and test model performance.
In addition to machine learning techniques to optimize regression modeling, this study examined spatial mismatch by converting kernel density raster maps into images for pattern comparisons. The Jaccard Similarity Index is a statistic applied to sample sets and a common tool in bioinformatics, image recognition, and text similarities. This dissertation applied data mining and cluster analysis to measure spatial inequality between unhealthy and healthy SNAP retailers, using the Jaccard Similarity Index to measure dissimilarity. This is the first known study that converts raster maps as images for pattern comparisons applying the Jaccard approach. Obesity and diabetes rates were significantly higher in census tracts with higher rates with African Americans and low socioeconomic status (i.e., SNAP participation). However, the relationship between the other racial/ethnic groups and health varied by region. Higher rates of Hispanics were related to higher diabetes rates in the Southwest and the Mid-Atlantic regions. However, higher rates of Asians were associated with higher diabetes rates in the West region, but an inverse relationship in the Southeast. In terms of spatial mismatch, distance to the nearest SNAP supermarket was not significantly associated with high obesity or diabetes rates, but the concentration of unhealthy SNAP retailers was significantly related to higher diabetes rates. However, there was a city-specific spatial mismatch between distance and high SNAP participation, but this cannot be generalized. Further, the spatial mismatch was strongest between high concentrations of unhealthy SNAP retailers and high SNAP participation, and high African American population rates. Food insecurity, including its health impact, has primarily been identified in public health research, and its research recognizes the need to improve healthy food accessibility and affordability through policy and environmental changes. This is where the planning discipline can address food insecurity and food accessibility and affordability. Integrating food systems in urban planning has been absent for some time but has been gaining attention in recent years. This study examined the relationship between local-level food access and federal policy. Planners can explore various strategies to increase the affordability and availability of healthy food options in low-income urban neighborhoods, such as financial incentives for retailers and urban agriculture policies. Since retailers apply to become SNAP retailers at the federal level, local planners can support small markets to meet the criteria to become SNAP vendors. The federal SNAP policy is an example of an individual-focused anti-poverty effort to increase purchasing power for food. Yet, the retailer siting influences purchasing behaviors, and geographic isolation and concentrated unhealthy retailers in low-income neighborhoods can inhibit access to affordable healthy options. Federal policies often focus on people– or place-based interventions to address poverty-related issues, yet both should be complementary.
Eichinger, Michelle, "Spatial Planning to Examine Regional Racial and Ethnic Disparities in Obesity and Diabetes by Analyzing the Supplemental Nutrition Assistance Program (SNAP) Retail Environment: Applying a Machine Learning Approach" (2021). All Dissertations. 2812.