A common approach to adjusting for seasonal and long-term trends is to use semiparametric models which incorporate a smooth function of time. The use of nonparametric smoothing in time series models of air pollution and health was suggested in Schwartz (1994a), where generalized additive Poisson models were used with LOESS smooths of time, temperature, dewpoint temperature and PM10. This approach can be thought of as regressing residuals from the smoothed dependent variable on residuals from the smoothed regressors. In this setting, the smooth function of time serves as a linear filter on the mortality and pollution series and removes any seasonal or long-term trends in the data
