FIGURE 7-2 The power that each study has to detect a given standardized regression coefficient.

levels and, in particular, on whether there are sufficient observations at high exposure levels to characterize the true shape of the dose response in that region. In fact, all three studies had fairly skewed exposure distributions, with a large number of subjects clustered at low exposure levels, along with a few subjects exposed at moderate to high levels. Such skewness in the observed exposure levels can be associated with other problems as well. For example, extreme observations have the potential to exert a strong influence on the results in such settings. Indeed, Crump et al. (1998) reported nonsignificant results from a regression analysis on all the children in the New Zealand cohort, but significant results after omission of a single child whose mother's hair Hg concentration was 86 ppm (4 times higher than that of the next highest exposure level in the study). We will see presently that dose-