Given a particular representation of f described in Section then choose the amount of smoothness to allow for f . We examine model selection approaches that have already been used extensively by investigators in the area of time series modelling of air pollution and health data. A general strategy is to use a data-driven method and to select the number of degrees of freedom(df) which optimizes a particular criterion. For example, one approach is to choose the df which leads to optimal prediction of the mortality outcome series (. Burnett and Gold- berg (2003)) and another is to select the df which best predicts the.
