unexplained drop in the number of daily deaths appears
in the third year of the study. The impact of these missing
values was examined by conducting additional analysis
using a data set restricted to the first two years of data.
Second, there were many days of missing values for PM10
during the study period, particularly during the rainy seasons.
Therefore, estimates of PM10 based on airport visibility
were developed for all four years of the study. By
using estimated values of PM10 and limiting the data to
include only the years with reliable daily mortality data,
we were able to test for the influence of missing or questionable
data. The results of these sensitivity analyses suggest
that data problems did not appreciably alter the overall
results. Bias does not appear to have been introduced
by using the estimated values of PM1