The purpose of this article is to investigate the sensitivity of food access models to a dataset's spatial distribution and the empirical definition of food access, which contributes to understanding the mixed findings of previous studies. Data was collected in the Dan River Region in the United States using a telephone survey for individual-level variables (n = 784) and a store audit for the location of food retailers and grocery store quality. Spatial scanning statistics assessed the spatial distribution of obesity and detected a cluster of grocery stores overlapping with a cluster of obesity centered on a grocery store suggesting that living closer to a grocery store increased the likelihood of obesity. Logistic regression further examined this relationship while controlling for demographic and other food environment variables. Similar to the cluster analysis results, increased distance to a grocery store significantly decreased the likelihood of obesity in the urban subsample (average marginal effects, AME = -0.09, p-value = 0.02). However, controlling for grocery store quality nullified these results (AME = -0.12, p-value = 0.354). Our findings suggest that measuring grocery store accessibility as the distance to the nearest grocery store captures variability in the spatial distribution of the health outcome of interest that may not reflect a causal relationship between the food environment and health.
Keywords: cluster analysis; food access; grocery store accessibility; obesity.