but it is equally difficult to justify the value as a representative or The value is easily discarded as a statistical outlier, valid result. It is emphasized and is thus discarded from subsequent evaluations. that even if it is a valid result, due to a hot particle, it does not characterize the site and it must be considered in conjunction with the other results from this site. Sample IV composite (5-10 cm) from Table 2 also appears to be unreasonably high. The result is probably due to cross-contamination and is discounted. Analytical Variability As part of the analytical effort, eight samples were analyzed in duplicate. These results are given in Appendix 2. The sample activities ranged from less than 5 to 970 fCi/g. Although the difference in vari- ability was not statistically significant, the ranges in results for the duplicates averaging greater than 100 fCi/g had less variability than samples below 100 fCi/g. The mean percent deviation between the duplicates (i.e., difference of the duplicates divided by the average of the two) was 40 percent. This is about twice the variability estimated by Krey and Hardy (1970) for results from 100-g aliquots of samples of the top 20 cm of soil. Both of these evaluations include the errors of obtaining a representative aliquot from a sample and sample analysis. A technique for expressing the variability of results, based on duplicates, in terms of the geometric standard deviation is presented by EPA, 1977c. This technique was used to estimate the coefficient of variation based on the assumption of a normal distribution, since the data sets in this report have been evaluated based on the assumed applicability of the normal distribution (see Appendix 2). The estimated coefficient of variation for these duplicates is 29 percent. Analysis of four duplicates for cesium-137 showed a significantly lower variability than that observed for plutonium-239. The average of the difference of the results divided by the mean was 9 percent (range of 0 to 18). Statistical Testing The various groups of data have been evaluated using several statistical tests. The limited amount of data minimizes the meaningfulness or power of these tests. Thus, emphasis is placed on the numerical trends of the data and the statistical tests are viewed as indicators of the uncertainties of the associated trends. There are insufficient data points in each set to evaluate the appropriateness of any specific statistical distribution. Although environmental data are often fit to a lognormal distribution (Denham and Waite, 1975), because of the small number of data points the normal distribution has been used to characterize these data. 655