ered. That study found associations with MeHg exposure in a population whose sources of MeHg exposure were similar to those in the Seychelles and used end points similar to those examined in the Seychelles. Although the New Zealand data have been available for some time, they have not been used extensively for risk assessment, possibly because until recently, they had not been subjected to standard peer-review procedures. A re-analysis of the New Zealand data by Crump et al. (1998), which underwent peer review, reported associations of prenatal MeHg exposure with several end points (when one extreme outlier was excluded), including four that were not found to be related to MeHg in the Seychelles study. The New Zealand study has been criticized for errors in matching exposed children to controls and for testing exposed children and controls at different ages (Myers et al. 1998). Those errors occurred in the 4-year follow-up but were corrected in the 6-year follow-up, which is the data set reviewed in this section. In addition, there is no information that would suggest the presence of differential measurement error across the studies. Any error of that type is likely to be nondifferential (i.e., unbiased), and it would reduce the likelihood of detecting associations between MeHg exposure and neurobehavioral test scores.
Data from the peer-reviewed pilot SCDS of 217 children assessed at 5.5 years (Myers et al. 1995) are also considered in this chapter. (Note that the nonstandard treatment of the data from the Revised Denver Developmental Screening Test (DDST-R) discussed in Chapter 5 was not an issue in the 5.5-year follow-up since the DDST-R was not given at that age.) Two of the four outcomes that were tested in both the pilot and the main Seychelles studies at 5.5 years of age were found to be associated with prenatal Hg exposure in the pilot study. The Seychelles investigators were cautious about drawing inferences from their pilot data, because the effects were substantially weaker when four outliers were excluded from the analyses and because socioenvironmental influences were not adequately assessed and controlled statistically. It is not clear, however, that it is appropriate statistically to exclude influential data points; many statisticians would instead recommend the use of data transformation to reduce their influence. Exclusion is appropriate only where a value appears biologically implausible (see discussion of the New Zealand outlier in the Benchmark Analysis section in Chapter 7). With regard to socioenvironmental influences, T.W. Clarkson