Combining Multiple Monitoring Data For Modeling Air Pollutant Concentrations

• Activity Number: 399
• Type: Contributed
• Date/Time: Thursday, August 15, 2002 : 10:30 AM to 12:20 PM
• Sponsor: Section on Statistics & the Environment*
• Author(s): Sandra McBride*+
• Affiliation(s): Duke University
• Keywords: Fieller-Creasy problem ; air pollution ; Bayesian semi-parametric model

Abstract

Studies of the association between airborne particulate matter (PM) and mortality have focused on the role of PM constituents, such as combustion-related components, soil particles, and secondary aerosols. For the Phoenix area, measurements and error estimates for PM constituents are available from multiple co-located monitors with varying degrees of temporal resolution. We develop a Bayesian semiparametric model for the true unknown mean of ambient soil concentrations. At monitors A and B, the unknown mean soil concentration can be represented as a linear combination of meteorological variables and radial basis functions. To account for bias in monitors relative to each other, we incorporate a multiplicative bias term in the mean for monitor B. Estimation of the bias term involves inference about the ratio of two normal means. We discuss the choice of priors for model parameters and modeling results, which allow information from multiple monitors to be combined and incorporated into a hierarchical model for PM concentrations.