Online Program

Controlling for Unobserved Spatially Correlated Confounders in Observational Studies

*Georgia Papadogeorgou, Harvard University 
Francesca Dominici, Harvard School of Public Health 
Corwin Zigler, Harvard School of Public Health 

Keywords: Spatial statistics, Observational studies, Causal inference

With the increasing amount of observational data available, appropriate methodology to correct for confounding in observational settings is of increasing interest and importance. Most methodology aims to identify important observed covariates and correct for confounding while relying on the assumption of no unobserved confounding. In the context of spatial data, we explore the notion of unobserved, spatially varying confounding and examine the importance of the scale of spatial correlation of the unobserved confounder on bias. We develop a new methodology that aims to eliminate bias in the estimation of effects that arises from the unobserved spatial confounders, making little or no assumptions on the structure of the confounding. We assess the performance of our proposed method via simulation studies under various scenarios and apply the procedure to data from the Acid Rain Control Program.