Abstract:
|
In air pollution cohort studies we cannot directly measure exposures for study subjects. Instead, ambient concentrations are predicted from data at different monitoring locations by means of land-use regression with spatial smoothing. The induced measurement error can be decomposed into components analogous to classical and Berkson error. We present two parallel treatments of this measurement error problem. In our parametric framework, we model the exposure surface using geostatistics, with a correctly specified mean model and a spatially correlated residual field. In our semi-parametric framework, we model the exposure surface as deterministic, but with a misspecified mean model and random monitor and subject locations. We argue that the semi-parametric analysis is more true to the underlying science, and then we discuss the consequences of analyzing the data within this framework, as compared to the parametric one. We illustrate the importance of our semi-parametric framework for understanding counter-intuitive results in an analysis of the association between exposure to fine particulate matter (PM2.5) and elevated systolic blood pressure in the NIEHS Sister Study cohort.
|
ASA Meetings Department
732 North Washington Street, Alexandria, VA 22314
(703) 684-1221 • meetings@amstat.org
Copyright © American Statistical Association.