AG 2007 10 15 - Laserfiche WebLink

Home Browse Search

Page 5-20: The rationale for selecting the 90/10, logistic/linear prior is not convincing. The completely uncertain shape of the true exposure response curve would make the 50/50 prior the natural choice. The analysis should then include a calculation of the posterior logistic/linear odds ratio to see to what extent the data prefer the latter. In fact, the Bayes factor, the ratio of the posterior to prior odds, is a common way of seeing to what degree the data do have a preference for one model over the other. Instead, the analysts have used the data and their impression of the logistic's superior fit to select the 90/10 even while conceding that the lack of data at lower exposure levels makes that choice tenuous. Page 5-29: Section 4.5.6 referred to in the first bullet near the bottom of the page is about meteorological data, not air quality, suggesting an error. CHAPTER 6 Page 6-31: The errata list should correct citations made on this page to Table 5-6 and 5-7. As well, the citation of Section 5.3.2.5 seems incorrect. Should it be 5.3.1.3? Page 6-21: Here we learn that the choice of the prior odds in favor of the logistic model over the linear one does make a difference to assessment of risk for all school children when comparing the more stringent standards to the current standard. Since the 50/50 would have seemed the more natural choice, the effect of this change should be clearly spelled out. Would it have affected the Staff s recommendations had it been chosen? Page 6-24: I agree with the last sentence on this page: the variability of the degree of protection afforded by the standards across urban areas does seem an important in evaluating the current standard. That point is revisited on Page 6-50. However, it never emerges how Staff incorporated that factor in their recommendations. Should standards be based on the worst case? ~~ ~~~. C-34