Controlling for localised spatio-temporal autocorrelation in long-term air pollution and health studies

摘要: Estimating the long-term health impact of air pollution using an ecological spatio-temporal study design is a challenging task, due to presence residual autocorrelation in counts after adjusting for covariate effects. This commonly modelled by set random effects represented Gaussian Markov field (GMRF) prior distribution, as part hierarchical Bayesian model. However, GMRF models typically assume are globally smooth space and time, thus likely be collinear any spatially temporally covariates such pollution. Such collinearity leads poor estimation performance estimated fixed effects, motivated this epidemiological problem, paper proposes new methodology allow localised smoothing. means that either geographically or adjacent allowed autocorrelated conditionally independent, which allows more flexible structures represented. increased flexibility results improved compared with global smoothing models, evidenced our simulation study. The then applied motivating investigating on respiratory ill Greater Glasgow, Scotland between 2007 2011.