Statistical analysis
The Distributed Lag Linear and Non-Linear Model (DLNM) was used to view
the direct and delayed effect of air pollution and environmental factors
on CE. Considering the distribution of number of daily CE events, the
Quasi-Poisson model is a more suitable model than the Poisson model.
Quasi-Poisson is usually used when the variance is greater than the
average, and this was the reason why we used this model. It was used in
combination with a DLNM to investigate the lagged and non-linear
effects. For all variables, the maximum lag day L was considered
up to 10 days.
The multivariate DLNM model for the air pollutants is:
\begin{equation}
\text{log\ E}\left(Y_{t}\right)=s\left({PM10}_{t};\eta_{\text{PM}10}\right)+s\left({NO2}_{t};\eta_{NO2}\right)+s\left({SO2}_{t};\eta_{SO2}\right)+s\left(\text{WS}_{t};\eta_{\text{WS}}\right)+s\left(\text{SS}_{t};\eta_{\text{SS}}\right)+ns\left(time,\left(4\times 6+4\right)\text{df}\right)+DOW+\sum_{k=1}^{6}{\gamma_{k}\mu_{\text{tk}}}\backslash n\nonumber \\
\end{equation}