Bias Analysis for the Use of Spatially Predicted Air Pollution Exposures in Linear Models of Air Pollution Health Effects

• Activity Number: 351
• Type: Contributed
• Date/Time: Tuesday, August 6, 2013 : 10:30 AM to 12:20 PM
• Sponsor: Section on Statistics and the Environment
• Author(s): Stacey Alexeeff*+ and Raymond J. Carroll and Brent A. Coull
• Companies: Harvard University and Texas A&M University and Harvard School of Public Health
• Keywords: spatial models ; air pollution ; kriging ; measurement error ; model uncertainty ; environmental epidemiology

Abstract

Land use regression models and kriging models are often used to model and predict air pollution exposures, and an underlying covariance structure for the air pollution surface must be assumed. Key sources of model uncertainty for predicted air pollution exposures include estimation error and model misspecification. We consider a linear model for air pollution health effects and present a thorough bias analysis for the use of spatially predicted air pollution exposures. In particular, we separately consider the issues of estimation error and model misspecification, and we find that a misspecified exposure model can induce significant bias in the effect estimate of air pollution on health. We propose a new correlated-errors SIMEX procedure and demonstrate that the procedure has good performance in correcting this bias. We also find that estimation error alone induces little bias, suggesting that model misspecification in exposures is the more important issue at hand. We demonstrate our methodology in an analysis of PM2.5 and birthweight in Boston. * Awarded 2nd place prize in JSM Section on Statistics and the Environment Student Paper Competition.