Feng, Cindy2020-11-032021-11-032020-092020-11-03Septemberhttp://hdl.handle.net/10388/13121In many environmental and epidemiological studies, the observed exposures are often highly correlated. Estimating the exposure-specific effects of the multiple correlated exposures on the health outcome in a statistical model can be challenging since multicollinearity among the exposures can lead to biased estimators. This study proposed a two-stage shared component model for addressing this challenge to utilize the information of the collected exposures fully. The first stage is a pollution model in which the shared and residual components are obtained to represent the common and unique effects from each correlated explanatory variable. The second stage is a disease model that the shared and residual components were included as explanatory variables for modelling the disease risk. The proposed model is motivated by an environmental health study that investigated the association between air pollutants and the respiratory hospital admissions in Greater Glasgow, Scotland, in 2011. The three highly correlated pollutants PM2.5, PM10 and NO2 are simultaneously modelled in the two-stage shared component model. Our results indicated that the air pollutants jointly increased the respiratory disease risk while NO2 has a stronger health effect. We also investigated the finite sample properties of the proposed two-stage shared component model, which demonstrated that the proposed method could help resolve the issue of multicollinearity with appropriate and easily interpretable coefficients.application/pdfAir pollutantsMulticollinearityShared component modelThesis2020-11-03