|
Abstract:
|
In environmental epidemiology, how the air quality change would affect human health is of great interest. Due to its universality and controversy, the air quality problem does not only affect long-term human welfare but also poses a major economical challenge to all the countries in the world. While investigating the air quality data, we often encounter the dilemma of multiple time series data with inner trend that cannot be accounted by the observed covariates. We have found that the current spline smoothing or kernel smoothing methods can produce positive or negative significant cross correlations depending on how the smoothing parameters are varied. Hence, three classes of robust and stable smoothers are proposed to de-trend the time series data. Their properties are demonstrated by theoretical justification via design of experiment on scenarios, as well as simulation that uses indistinguishable datasets from the original. Finally, general guidelines are provided on how to reveal abnormal signals from time series data with inner trend.
|
