Stochastic microenvironment models for air pollution exposure.

Exposure assessment is a crucial link in air pollution risk assessment and management. With the recent advances in instrumentation, it has become possible to measure air pollution exposures in the vicinity of the individual human subjects, using either personal monitoring or microenvironment monitoring. For many important pollutants such as CO, NO2, and VOC, the air pollution exposure depends crucially on the location and activity of the individual: indoor versus outdoor, smoking versus not smoking, etc. The stochastic microenvironment models were developed to relate air pollution exposure to the location and activity. We review the two major existing models, the Cartesianization method (Duan, 1980, 1982, 1987) and SHAPE (Ott, 1981, 1982, 1984), and compare their assumptions and implications. We also propose a new model, the variance components model, which includes both Cartesianization and SHAPE as special cases. The variance components model considers both long-term average concentrations and short-term fluctuations. The Cartesianization focuses on long-term averages, while SHAPE focuses on short-term fluctuations. We propose to choose among the three models by examining the variance function which relates variability to averaging time. The theory is applied to the data collected from U.S. EPA's Washington CO Study, with the variance function estimated using Carroll and Ruppert's (1984) transform-both-sides regression model and Duan's (1983) smearing estimate. For the microenvironment in transit, both long-term averages and short-term fluctuations are important.