Evaluating statistical appropriateness of alternative regression techniques for estimating ambient nitrogen dioxide concentrations

In order to protect public health, the US Congress passed the CAA which requires estimating ambient air pollutant concentrations using models. The current regulatory model can be difficult to use when estimating air pollution exposures in large urban population centers. Land Use Regression (LUR) mod...

Full description

Saved in:
Bibliographic Details
Main Author Gunter, James Thomas
Format Dissertation
LanguageEnglish
Published ProQuest Dissertations & Theses 01.01.2011
Subjects
Environmental Health
Occupational health
Occupational safety
Online AccessGet full text

Cover

More Information
Summary:In order to protect public health, the US Congress passed the CAA which requires estimating ambient air pollutant concentrations using models. The current regulatory model can be difficult to use when estimating air pollution exposures in large urban population centers. Land Use Regression (LUR) models offer a reasonable alternative. The central research question of this study is “Are LUR models that consider correlation among observations statistically more appropriate than are LUR models that assume independence?” Information is obtained from the US EPA, Oklahoma Departments of Transportation and Environmental Quality, Oklahoma Mesonet, National Climatic Data Center, and National Oceanic and Atmospheric Administration. AREMET is used to determine boundary layer conditions. The analytical dataset is generated from this information that includes meteorological, point and mobile source, and temporal measures. LUR models are specified that use mixing height, stability, hourly traffic count, count of major point sources, and season to estimated hourly NO 2 concentrations at three state and local air monitoring stations. Identical models are estimated using OLS and mixed model regression. OLS estimates only population effects and assumes independence among observations. Mixed models estimate fixed and random effects and consider correlation among observations. Using mixed models, the first order autoregressive covariance structure best fits the data. The results of this study lead indicate that autocorrelation should be accounted for when estimating LUR models so that the significance of predictive values can be accurately assessed. The mixed LUR model was more precise and had more predictive power than the OLS LUR model while both provided unbiased coefficients estimates.
ISBN:1267512873
9781267512871