Modeling multiple pollutants at multiple sites, with application to acute
|Aim of research: To develop
models for pollution measures taken at multiple sites.
Data: The primary data are on seven pollutants from 23 sites in London
over the period 1981--1997. Meteorological data are also available. The
secondary data set is on PM$_10$ only from 15 sites in the greater
Seattle area over the period 1990-1995.
Methods: The pollutants will be modeled using Bayesian dynamic
generalized linear models. Specifically the framework of state space
models will be used with the observed pollution level being modeled as
arising from some true underlying level, but corrupted with measurement
error. These true underlying levels are assumed to have structure both
in space and time. Under this framework, missing data will be
automatically imputed which, in particular, is useful for the analysis
of health studies.