homogeneity. The suggestion of both ACPA and the Agency to stratify on vulnerability makes sense, because it can help to assure that more attention is paid to CWSs most likely to have relatively high mean concentrations, which is the part of the distribution that the Agency is most interested in. But again, it is not clear that this stratification will produce homogeneous groups, or increase statistical efficiency. More attention needs to be paid to stratification as a means of variance reduction, and to the complex and sometimes ambiguous relationship between sub-domains and strata.
Regarding the preferential selection of CWSs with multiple pesticides, in the Agency background document, the Agency reported that a full suite of analytes would be sampled at each CWS in the sample design, but that the design would be tailored to produce reliable estimates only for the targeted pesticides. The notion behind this question is that if just the right CWSs were selected, the number of pesticides with reliable estimates of exposure could be increased. While it may be possible to weigh the CWSs somewhat, it is critical that the design remain probabilistic. In the end, the CWSs chosen must represent the set not chosen, and there is the danger that with preferential selection you might draw a set of CWSs that are too unique. Instead of this approach, perhaps after the sample is selected it could be examined for cases where a few more samples in a pesticide use area would produce reliable estimates for that pesticide, and a second sample drawn to add the few additional CWSs needed to extend the list of pesticides for which reliable estimates could be obtained.
3.) EFED is considering stratifying the CWSs by vulnerability. The estimation of vulnerability would attempt to identify, for each pesticide and CWS, the likelihood of contamination. EFED would then randomly select CWSs from the vulnerability strata for inclusion in the survey.
How should we combine factors into an estimate of site vulnerability?
Would additional or different strata increase the usefulness of the data in addressing
the objectives of the monitoring study?
What are the relative merits of using vulnerability-based as opposed to
geographically-based stratification methods?
The multiple pesticides of interest, the variability of hydrological, climatological, water body type features of individual sites present a very complex problem in stratification. Hence, this response has multiple components. Optimal stratification for a sample of CWS is relatively straightforward for a single analysis problem (a single pesticide, a single statistic of interest); however, the multiple objectives of this study preclude a choice of a single optimum for all questions of interest. For this reason, it is advisable that the Agency establish a prioritization of pesticides to study in depth based on grounds of scientific importance in the assessment of health risks. The SAP recognizes that the Agency, industry, and the public have concern over potentially hundreds of different pesticide compounds, but to achieve useful scientific and regulatory data from a study of this scope, hard choices over priorities must be made.