International Day Of Women in Statistics and Data Science

Quantifying neighborhood food environments and understanding their relationships with residents’ health is a public health priority. Using simple, error-prone food access metrics (like the shortest straight-line routes to healthy food stores) introduces measurement error and biases downstream statistical models, but measuring the more-accurate, map-based ones (like shortest driving routes) for entire studies is often implausible. Fortunately, adopting a two-phase design can harness the best of both metrics by combining the error-prone access measures for the entire study and the more-accurate ones for a chosen subset in a partial validation study. This validated subset can be strategically chosen to not only reduce bias but further improve efficiency when modeling relationships between health and the food environment. Technically, any information that is fully available for all neighborhoods can guide the validation sampling strategy. One such promising design paired stratification with Neyman allocation and sampled based on the influence function within each stratum. Using simulations and data for the Piedmont Triad Region of North Carolina, various validation sampling designs were evaluated to quantify the associations of diabetes count and obesity prevalence with neighborhood-level access to healthy foods, fitting two separate Poisson regression models, one for each outcome. We assess which design suits each model, and whether any is robust across settings.