Nonparametric modeling of the effects of air pollution on public health

Abstract

This thesis aims to analyze the effects of air pollution on public health across 15 populous cities in USA based on daily observations from 1987 to 1998. In the analysis, we firstly perform Efficient Dimension Reduction (EDR) to reduce the complexity of the data. Then we compare the cross-validatory (CV) values, which assess models with their forecasting performance, of a Generalized Additive Model (GAM) and a general nonparametric regression model. The model with smaller CV-values is preferred. Finally, we discuss whether the commonly used GAM is acceptable to quantify the effects of pollution. Our results show that air pollutants (PM10, O3, SO2, NO2 and CO) at current levels, acting with temperature and humidity together, have adverse effects on public health. The influential hazards are O3, PM10, and weather variates. For model selection, our results suggest that EDR is necessary and that the general multivariate model incorporating EDR outperforms GAMs.