Analyses indicated that source types were a significant predictor of RR, whereas apportionment group differences were not.
Instead, the same pre-existing reference PM mass and constituent data sets from two cities (Washington, DC, and Phoenix, AZ) were sent to various leading source apportionment research groups in advance of the workshop (in December 2002), and each group individually analyzed the same data sets for daily source PM2.5 contributions.
With regard to the PM2.5 mass apportionments, the findings of this intercomparison among results from some of the leading source apportionment research groups indicate that the same major source types (those that contribute most of the PM2.5 mass at each site), with similar elemental makeups (i.e., key tracers), are consistently identified by different groups in each city.
Overall, the results of this intercomparison of the health effects apportionments found that variations in PM source apportionment research group or method introduced relatively little uncertainty into the evaluation of differences in PM toxicity on a source-specific basis, adding an average of only approximately 15% to the overall source-specific mortality RR uncertainties.
Finally, we evaluated the size and significance of the additional variability introduced to the PM mortality, time-series analysis by variations in the source apportionment process across groups and methods, consistent with the primary goal of this workshop.
The same Phoenix base mortality model was applied to source apportionment analyses of all groups to provide a consistent basis for comparison across source components and groups (i.e., to eliminate model specification variability from the analysis).
The source apportionment results for each group were combined with the mortality data in Washington and Phoenix, and time-series mortality regressions were then run (Ito et al., in press; Mar et al., in press).
If a constant relationship existed between these groupings, apportionment would not be difficult.
This approach provided directly comparable mortality effect estimates for each source category and for apportionment modeling results of participating groups.
This result indicates that variations in choice of research group or source apportionment method have only a small effect on variations in the RR estimates, relative to the variations in RR caused by different source components.
