Estimating Health Effects of Particulate Matter Sources in the Presence of Censored Air Pollution Concentrations

• Activity Number: 592
• Type: Topic Contributed
• Date/Time: Wednesday, August 7, 2013 : 2:00 PM to 3:50 PM
• Sponsor: Section on Statistics and the Environment
• Author(s): Jenna Krall*+ and Roger D. Peng
• Companies: Johns Hopkins University and Johns Hopkins University
• Keywords: particulate matter sources ; health effects ; censoring ; factor analysis

Abstract

Identifying sources of particulate matter (PM) air pollution most harmful to human health would allow more targeted regulation of PM. Source concentrations are not directly collected and are inferred from chemical constituent concentrations collected from ambient PM monitors. Daily constituent concentrations sometimes fall below the monitor's analytical minimum detection limit (MDL) and there is not agreement on how missing data should be imputed. In a simulation study, we examine bias in source-specific health effects under common imputation methods. To impute concentrations below the MDL, we propose a Gibbs sampling approach, which leverages correlation among PM chemical constituents expected from shared sources. The amount of bias in health effects varies by the imputation method applied and different methods perform best depending on the amount of missingness. We apply these methods to PM chemical constituent data and non-accidental mortality data in the U.S. and demonstrate that differences exist in mortality risk estimates by imputation method. PM constituent concentrations below monitor detection limits can bias health effect estimation of PM sources.