We applied natural smooth (ns) functions of calendar time with 7 degrees of
freedom (df) per year to exclude unmeasured long-term and seasonal trends in the
time-series dataset (Peng et al., 2009). We incorporated the ns functions of mean
temperature (6 df for the period) and relative humidity (3 df for the period) to adjust
for the potential nonlinear confounding effects of weather conditions (Chen et al.,
2012). We also included the day of the week as an indicator variable in the basic
models. After establishing the basic model, we introduced the air pollutant concentrations
into the single-pollutant model one at a time to estimate their associations
with asthmatic hospitalization.