AAAR 31st Annual Conference
October 8-12, 2012
Hyatt Regency Minneapolis
Minneapolis, Minnesota, USA
Abstract View
Estimating Health Effects of Air Pollutants in Pittsburgh from 2001-2002 Using Autoregressive Moving Average (ARMA) Time Series Structural Equation Models (SEMs)
RICHARD BILONICK, Daniel Connell, Evelyn Talbott, Judith Rager, Lynne Pavlic Marshall, University of Pittsburgh
Abstract Number: 378 Working Group: Instrumentation and Methods
Abstract Statistical modeling has been used extensively to estimate the effects of air pollutants on human health. Single-pollutant models have commonly been used where only one pollutant of interest at a time is modeled (usually adjusting for weather and for day-of-week, seasonal and other trends), even though it has been known since Fisher's work in the 1920s that this approach tends to introduce bias because it does not account for other pollutants with significant effects. More recently, multi-pollutant models have been used to overcome this limitation, in spite of problems with collinearity. Most importantly, however, neither the single-pollutant nor the multi-pollutant approach accounts for the typically significant amount of measurement error in air pollution measurements, which further biases model coefficients in unpredictable ways. Simply including additional pollutants in regression models or even time series models (regression models with autoregressive and/or moving-average terms) is unlikely to provide a clear understanding of which pollutants are most important. Rather, we propose the use of structural equation models (SEMs) to appropriately account for the statistical complexities associated with estimating the health effects of multiple air pollutants. SEMs provide a comprehensive and rich set of tools for producing models that can capture complex interrelationships among explanatory factors and their relationships with the response (structure) and simultaneously account for autocorrelated measurement error, autoregressive (lagged and cross-lagged) effects, and irregularly timed and/or missing measurements. Additionally, SEMs which include observed explanatory variables as indicators of latent explanatory factors can avoid problems with multicollinearity. The results of using SEMs to estimate health effects of air pollutants, including chemical constituents of PM2.5, in Pittsburgh, Pennsylvania, during 2001-2002 will be presented using data from the Pittsburgh Aerosol Research and Inhalation Epidemiology Study (PARIES).