Title: Spatial and spatio-temporal confounding in biometrical applications: ubiquitous undesirable effects and alleviation proposals

4 August 2020 | 13:00 UTC/GMT

Time Zone Converter

Speaker Abstracts 


Spatially referenced data are very common in biometrical applications, but in spite of the huge amount of research in the field, there are still many important issues that require further research and attention. Spatial regression models are widely used to deal with spatial data and they typically include spatially correlated random effects in addition to covariates. In problems involving spatial data, it is the norm, rather than the exception, that covariates of interest exhibit spatial patterns. The impossibility of distinguishing the covariate effects from 
spatially correlated random effects or error terms is known as spatial confounding, a major issue that has received attention but still requires further insight. Some of the issues with spatial confounding arise also from unmeasured spatial confounders.

Different undesired effects of spatial confounding have been documented in the literature showing that ignoring spatial confounding can lead to counterintuitive results and wrong conclusions. One of them is that the fixed effects estimates are biased and their variances are inflated. This has the dramatic consequence that potential associations between the phenomenon of interest and the covariates are blurred. Spatial confounding can also have effects on spatial prediction and interpolation: if the covariate and the random effects are correlated the spatial predictor is biased though the mean squared prediction error may be reduced.

Lola Ugarte, Public University of Navarre

Janine Illian, University of Glasgow
Careful prior selection and cautious inference for spatial point processes

Tomas Goicoa, Public University of Navarre
Assessing the association between socio-demographic covariates and dowry deaths in Uttar Pradesh

Patrick Schnell, The Ohio State University
Mitigating bias from unobserved spatial confounders using linear mixed models

Marcos Prates, Universidade Federal de Minas Gerais
Projection-based approach to alleviating Spatial Confounding in Spatial Frailty models

Georgia Papadogeorgou, Duke University