Abstract: Epidemiological analyses of environmental risk factors often include spatially-varying exposures and responses. In this context, any unmeasured, spatially-varying factor can lead to spatial confounding. This phenomenon can introduce bias in estimates of associations with the response. In this talk, I present how thin-plate regression splines (TPRS) can be used to mitigate spatial confounding bias. I first introduce the concept of spatial splines and show how individual splines are combined in an analysis to model the spatial variability. Next, I introduce the effective bandwidth of a spatial basis as a measure to interpret the spatial scale of a TPRS basis. I then present several current approaches that adjust for spatial confounding via TPRS and introduce a new hybrid approach that combines features of existing approaches. These methods differ in how they select the number of splines to include in the spatial basis and which models include the spatial basis. I compare these methods in a simulation study to make a recommendation for the best approach to mitigate spatial confounding bias. Finally, I conclude with a description of my current work extending the presented methodologies into a spatiotemporal framework and give examples of how undergraduate students can be involved in my research moving forward.
