The methods we will use to estimate whether there have been changes over
time in the effect of PM10 on mortality are similar to those used by the US investigation
described earlier (Dominici et al., 2007). Overdispersed Poisson log-linear
models taking the following form: