Bias and Variance Corrections for Spatially Varying Measurement Error

• Activity Number: 230
• Type: Topic Contributed
• Date/Time: Monday, July 30, 2012 : 2:00 PM to 3:50 PM
• Sponsor: Section on Statistics and the Environment
• Author(s): Stacey Alexeeff*+ and Brent Coull and Raymond Carroll
• Companies: Harvard School of Public Health and Harvard School of Public Health and Texas A&M University
• Keywords: measurement error ; air pollution ; spatial ; epidemiology

Abstract

Spatio-temporal modeling of air-pollution levels has become widespread in air pollution epidemiology research as a way to improve exposure assessment. However, a health effect analysis that uses predicted values from an exposure model results in exposure with measurement error, and the magnitude of the error may vary by location. Statistical methodology should properly account for the uncertainty associated with modeled predictions. We explore simulation based approaches, focusing on functional models which place minimal assumptions on the distribution of the exposures. These approaches provide a more flexible correction method that could be applied to many different exposure prediction models. One limitation of existing investigations is the use of convenient representation of the exposure surface that may or may not be realistic in practice, including the use of very smooth exposure surfaces to represent the true exposure in simulation studies. To generate a more realistic exposure surface, we use air pollution data collected by satellite on a fine scale grid over the greater Boston region. We apply the proposed methods to an analysis of air pollution and birthweight in Boston.