APPENDIX 3 ROCKY FLATS REPORT INTRODUCTION The following discussion presents assumptions and methodology used to evaluate the plutonium content of soil around the Rocky Flats Plant (RFP), Colorado. An important issue which this analysis hopefully will begin to define involves determining which sampling methodology has less variability. A second question is whether plot or subsample variability is of the same order as that observed over the entire sample site. Although data for plutonium in soil have been observed to have a log~ normal distribution, no transformation was made in this study. The number of samples per sample location are five or less and Denham and Waite (1975) note, for about 10 or less observations, no specific population distribution is superior. The sample sites were subdivided into sample plots which were assigned at random to collecting parties. Special collections were made on selected plots which are reflected as subsamples in subsequent discussions. For each data set, the following parameters were calculated: mean, standard deviation, variance, and coefficient of variation. The sample design was based on four distinctly different pedological and geomorphological sites at varying distances and bearings from the RFP. Therefore, the relative contribution of RFP plutonium versus worldwide fallout is variable; and it is important, therefore, to keep inference about a specific block applicable to just that sampling block. Table 1, in the text, gives a list of plutonium results and statistics for each data set. The two methods of soil sampling were for the upper l-cm and 5-cm intervals. The means of the 5-cm sample results appear to be inherently larger than those for the l-cm samples and the variances appear to be correspondingly larger as well. Before comparing variances on plutonium concentrations for the 1- and 5-cm samples, the results were pooled and a Pearson product correlation coefficient was calculated to test whether mean values and variances were independent (Koch and Link, 197la). The coefficient computed, 0.913, was significant at the 95 percent confidence level (t = 5.482, df = 6). The dependence of the variances on the means can also be shown by a linear plot of these values. Therefore, the variances are not independent of the sample means and the direct comparison of variances is inappropriate in assessing technique variability. In order to compensate for variance dependency on the mean between the l-cm and 5~cm sampling procedure, the sample coefficient of variation 676