While time series analyses have demonstrated that airborne particles are associated with early death, they have not clarified how much the deaths are advanced. If all of the pollution-related deaths were advanced by only a few days, one would expect little association between weekly averages of air pollution and daily deaths. The author used the STL algorithm to classify data on air pollution, daily deaths, and weather from Boston, Massachusetts (1979-1986) into three time series: one reflecting seasonal and longer fluctuations, one reflecting short term fluctuations, and one reflecting intermediate patterns. By varying the cutoff point between short term and intermediate term, it was possible to examine harvesting on different time scales. For chronic obstructive pulmonary disease, there was evidence that most of the mortality was displaced by only a few months. For pneumonia, heart attacks, and all-cause mortality, the effect size increased with longer time scales. The percentage increase in all deaths associated with a 10-microg/m3 increase in PM2.5 rose from 2.1% (95% confidence interval: 1.5, 4.3) to 3.75% (95% confidence interval: 3.2, 4.3) as the focus moved from daily patterns to monthly patterns. This is consistent with the larger effect seen in prospective cohort studies, rather than harvesting's playing a major role.