Title : Preferential sampling of environmental monitoring networks for use in downstream causal analysis of adverse birth outcomes: A Bayesian framework

Preferential sampling can occur in environmental monitoring networks because of financial means, city planning, ease of access, or population density within the region of interest. Selective monitor placement within a network may introduce systemic bias into spatial methods used to map total pollution surfaces. Here, a hierarchical Bayesian framework is used to jointly model the observation and site-selection processes through shared latent effect parameters, spatial processes, independent site variation, and monitoring network fluctuation over time of sulfur dioxide and ozone measured in South Dakota (2011-2021). Interpolated pollution surfaces will be used in downstream causal analysis of adverse birth outcomes where levels of exposure are assigned based on residential county. The observation process captures global temporal trends and smooth spatially correlated variation whereas the site-selection process captures changes in the log-odds of site-selection over time. After controlling for site-selection variations, results indicate that where sulfur dioxide and ozone measurements were taken is positively associated with the measurement itself. Fluctuation in the monitoring networks over time was not found to be influenced by clustering of observations. The estimated log-odds change in retention of established monitors was near zero, indicating that locations were maintained over time.

Emma Brookman is a PhD student at South Dakota State University where she has completed her undergraduate and master's degrees. She is originally from Rochester, Minnesota. Environmental epidemiology has piqued her interest during her studies because of the range of actionable results and insights derived from this methodology. Investigating the association between air and water pollution and preterm birth was the topic of her master's thesis. She will be extending this research to encompass causal inference and environmental epidemiology with applications to environmental policy and public health during her PhD studies.