B0556
Title: A Bayesian perspective on spatial+
Authors: Isa Marques - The Ohio State University (United States) [presenting]
Paul Wiemann - The Ohio State University (United States)
Abstract: Spatial models are used in a variety of research areas, such as environmental sciences, epidemiology, or physics. A common phenomenon in such spatial regression models is spatial confounding. This phenomenon is observed when spatially indexed covariates modeling the mean of the response are correlated with a spatial random effect included in the model, for example, as a proxy of unobserved spatial confounders. Several solutions to spatial confounding have been brought forward, including the method Spatial+ (Dupont et al., 2021). Spatial+ is based on a two-stage frequentist model. One notably absent point is the consideration of the additional uncertainty arising from the first stage estimation determining the residuals. Incorporating uncertainty can be achieved via a structural equation model. While similar point estimates will result from either method, a structural equation model has the distinct advantage of integrating all steps in one joint optimization problem that can be easily embedded in a Bayesian framework. We evaluate the performance of Spatial+ in a Bayesian framework, with special attention to uncertainty propagation. Both simulated and real datasets are considered.
