Statistical modelling of nonlinear long-term cumulative effects

Abstract

In epidemiology, bio-environmental research, and many other scientific areas, the possible long-term cumulative effect of certain factors has been well acknowledged, air pollution on public health, exposure to radiation as a possible cause of cancer, among others. However, there is no known statistical method to model these effects. To fill this gap, we propose a semi-parametric time series model, called the functional additive cumulative time series (FACTS) model, and investigate its statistical properties. We develop an estimation procedure that combines the advantages of kernel smoothing and polynomial spline smoothing. As two case studies, we analyze the effect of air pollutants on respiratory diseases in Hong Kong, and human immunity against influenza in France. Based on the results, some important issues in epidemiology are addressed.