Modelling Health Risk Due to Air Pollution Using Linear Filter Theory

Abstract

The relationship between human health and air pollution has been an area of active research for several decades. Of particular interest is the ability to estimate the health risk due to particular pollutants. Normally a model between mortality (or morbidity) and pollutants such as ozone, NO2, and fine particulate matter is estimated; with the model coefficients representing the risk due to its associated pollutant. Often a general additive model (GAM) framework is used and can account for lagged effects of up to a few days. We present a model for risk based on linear filter theory. The result is a linear finite impulse response filter associated with each pollutant and a measure of total risk which is represented as a function of these individual filters. This non-parametric model allows for much longer lag effects to be considered and deals naturally with the inherent colinearity amongst lagged variables.