Weighing the Distance Factor:
A spatial analysis of the association between accessibility to healthy food stores and obesity rates in Louisiana
Background
Obesity
Globally obesity rates have increased and have almost doubled since 1990. According to the World Health Organization (WHO), more than 890 million adults were classified as obese in 2022.
A report from the Centers for Disease Control and Prevention showed that 42.4% adults were obese in 2017-2018 in the United States (Hales et al., 2020).
Obesity is linked to multiple health related complications including diabetes, cardiovascular, neurological diseases, and chronic respiratory diseases and digestive issues (WHO, 2018).
Obesity is a multidimensional probelm
It is a product of these factors:
GENETIC * BEHAVIOURAL* ENVIRONMENTAL*SOCIO-ECONOMIC
Obesity and Food Environments
Research consistently demonstrates that access to abundance of unhealthy food options which are usually termed as food swamps or access to the lack of healthy food options termed as food deserts significantly impact public health (Aretz et al., 2023; Cerceo et al., 2023).
Finkelstein et al. (2005) explored the economic factors behind high obesity rates.
Rationale of the Study
The comprehensive review of concepts related to obesity, food access and desert and the methods utilized showed that there is a significant association between obesity rates and dietary choices as well as physical activity.
Null hypothesis (H 0 ): There is no association between accessibility to healthy food options and obesity.
Alternative hypothesis (H a ): There is association between accessibility to healthy food options and obesity.
Study Area
Why Bossier and Caddo Parish?
- Louisiana ranks seventh for the highest obesity rates in the United States, with Shreveport ranked as the fifth fattest city in 2021 (Eschette, 2023; Ferrel, 2021).
- Several factors contribute to these high obesity rates, including low consumption of fresh fruits and vegetables due to limited accessibility and a lack of physical activity (Joshu, 2024).
- Additionally, high poverty rates force residents to consume “preservative-rich, empty-calorie foods” (Bristol, 2023).
Data
Data | Type | Source |
|---|---|---|
Location of Healthy Food Stores | Vector: Point Data | Safe Graph (ArcGIS Pro: Business Analyst); supermarkets and grocery stores;warehouse clubs and supercentres and fruit/vegetable markets (North American Industry Classification System Code) U.S. Department of Agriculture, Food and Nutrition Service, 2023 |
Road Network | Vector: Multiline | Open Street Map |
Census Tract Boundary | Vector: Polygon | U.S. Census Bureau, 2023 |
Adult Obesity Rate | Tabular | Centers for Disease Control and Prevention, 2023 |
% of Adults with Low Physical Activity | Tabular | Centers for Disease Control and Prevention, 2023 |
Total Population | Tabular | U.S. Census Bureau, 2022 |
Other Socio-Economic Variables: 1. % With cash public assistance or Food Stamps/SNAP 2. % Unemployed (16 years and above) 3. % Non-white Population 4. % Less than high school degree 5. Median Household Income (USD) 6. % of Population below Poverty Line 7.% of Population with no vehicle access | Tabular | U.S. Census Bureau, 2022 |
Datasets Used
Methods
Preprocessing for Stores
- For the SNAP retail location dataset, the point data included convenience stores. That category of stores was excluded.
- This data was then merged with the Safe Graph’s data from ESRI’s business analyst.
- Merging the data resulted in duplication of several points, hence one point for each location was retained.
- This gave a total of 132 stores selling healthy food, mainly clustered across Shreveport and Bossier City
Preprocessing for Network
- The network dataset was reprojected to NAD 1983 (2011) StatePlane Louisiana North FIPS 1701 (US Feet) coordinate system.
- The travel time was calculated by dividing the length of each road segment by the speed of that road segment.
- This was used to create the network dataset layer, for which travel time was used as the impedance and length was the distance cost.
Preprocessing for Tabular Data
The tabular datasets were extracted from their respective larger datasets and then organized and merged into one table using Microsoft Excel and R programming.
Incase of missing data, the median value of that variable was imputed for the missing cases, as the median is more likely to represent the entire distribution than the mean value.
Calculating the Accessibility Index
Steps for Two Step Floating Catchment Area Method
Results and Findings
- The two step floating catchment area method showed varied patterns of accessibility for the three-time thresholds considered in this study.
- The spatial pattern of accessibility for the ten-minute threshold revealed that accessibility was high to very high in the central part of the study area. As distance increased the accessibility for this threshold decreased.
- The spatial pattern of accessibility for the twenty-minute time threshold was similar to the previous pattern of accessibility.
- However, the tracts falling in the central location which had higher values of accessibility now have relatively lower values.
- Some of the tracts which had very low accessibility now have low accessibility.
For the thirty minutes time-threshold accessibility improved the most for all the tracts.
Distribution of Census Tracts by Accessibility Index to Healthy Food Stores for Different Travel Time Thresholds
- Accessibility improved as the time interval increased for the very low and low accessibility categories as the number of census tracts falling in these categories decreased.
- The number of census tracts with moderate accessibility categories increased with the increase in time thresholds.
- In contrast, the number of census tracts with high and very high accessibility decreased with increase in travel time.
Regression Diagnostics and Coefficients for Accessibility to Healthy Food Stores (left); Regression Diagnostics and Coefficients for Spatial Lag Model of Obesity and Accessibility to Healthy Food Stores (right)
- The multivariate regression revealed that accessibility and obesity are negatively associated. In other words, for every one unit increase in accessibility score the obesity rates among adults decrease by 0.44 percentage points holding all else constant.
- The results also showed that low physical activity is directly related to obesity rates. With one percentage point increase in low physical activity obesity rates increases by 0.77 percentage points, holding all else constant.
- The statistically significant and positive value of the spatial lagged dependent variable indicates that obesity rates are spatially autocorrelated.
Conclusion
The study rejects the null hypothesis, as the p-value for accessibility index is 0.036 which was less than the significance level of 0.05.
The analysis of accessibility index to healthy food stores revealed that the degree of accessibility changed with respect to different time thresholds.
Accessibility to healthy food stores which was used as a measure of access to food environments showed statistically significant relationship with adult obesity rates.
The findings and results highlight and re-emphasize that obesity is not a product of one factor.
In lines with the analysis, there is a clear cut need to establish more healthy food store options in the peripheral parts of the study area.
The coefficients of the control variables like percentage non-white and percentage with less than high school degree also showed positive association with obesity rates. This highlights the importance of greater targeted interventions for certain demographic groups.
Limitations
- The study considered three driving time thresholds: 10, 20 and 30 minutes. The results of the study might change if the time thresholds are changed.
- Only one mode of transportation is considered, results might be different if different modes like walking and transit are also considered.
- The two step floating catchment area method considered the number of stores, but the store quality and factors like food prices was not incorporated. Qualitative analysis could be done further to take this into account in future.
