This thesis develops statistical methodology for an important area of environmental epidemiology, that of the relationship between short-term exposure to air pollution and mortality or morbidity which has been a public health concern for over fifty years. The majority of studies investigating this relationship are based on ecological data, and estimate a group level association between ambient pollution levels and population aggregated mortality. This association is typically estimated with Poisson regression models, which make a number of simplifying assumptions about the underlying processes that generate the data. The work presented in this thesis extends the standard approaches to modelling these data in three main ways, the first proposing the use of autoregressive processes rather than smooth functions to remove any long-term trends and temporal correlation in the daily mortality series. The second extension relates to the pollution-mortality relationship, and investigates whether it changes over time rather than being constant or a dose-response curve. The remainder of this thesis investigates the importance of correctly estimating pollution exposures, and how mis-estimating them affects the resulting health risk. These extensions are implemented using Bayesian hierarchical models with estimation achieved via Markov chain monte carlo simulation. For the first two extensions likelihood based alternatives are also presented, using a combination of maximum likelihood and least squares methods. The thesis ends with a concluding discussion.
| Identifer | oai:union.ndltd.org:bl.uk/oai:ethos.bl.uk:439177 |
| Date | January 2007 |
| Creators | Lee, Duncan Paul |
| Publisher | University of Bath |
| Source Sets | Ethos UK |
| Detected Language | English |
| Type | Electronic Thesis or Dissertation |
Page generated in 0.1464 seconds