The quality of the data analyzed was also affected strongly by the various problems encountered in taking samples in the field. For example, the surface soil sampling design was quite complex to execute in the field, and it took time for a new sampling crew to learn to take these samples properly. Also, the equipment had a tendency to deteriorate or be altered inadvertently when parts were replaced, so that later data may not have represented the same population as earlier data. The primary result of these and similar field problems was to increase the sample variance, making comparisons among data sets more difficult. Data quality was unavoidably altered to an unknown extent by the engineering operations that were necessary to allow data collection. For instance, if the vegetation were removed totally, as on Janet, the resulting soil disturbance altered the distribution of the TRU activity in the soil. If only aecess lanes were cut, as on other islands, soil disturbance was reduc.Jj but not eliminated. In addition, the data had to be corrected for signal attenuation from the remaining brush, using a subjective estimate of the amount of brush and an empirical brush correction factor. Because of these factors, the general principle used for choosing between alternative statistical approaches was to use the simplest method that would do the job. Certain types of data that were reported by others to the statistician were accepted as accurate because there was no way to verify the information. Examples are the total volumes of soil removed, the nominal depth of soil profile samples taken where the surface was uneven, actual boundaries of soil lifts, brush cover estimates, and similar information. No estimates of variance or reliability could be made for such data, so they were accepted at face value. Cleanup Criteria. The cleanup criteria were stated as averages over specified areas such as 0.25 ha, and specified depth intervals such as 0-3 em. Therefore the statistical methods used had to be appropriate for making estimates of area averages for a given depth interval. Also, the criteria required that the estimation error be considered, so an estimate of the error also had to be made. However, it was not clear at the beginning of the project whether the criteria applied to upper bounds or lower bounds on the estimates. The conservative approach of applying the criteria to the upper bounds was actually used, that is, soil was removed if the estimate plus half its standard deviation exceeded the applicable criterion. The subsurface cleanup criterion was difficult to interpret. Eventually the criterion was restated to reflect the limitations of the subsurface data, so the statistical analysis could aim at locating boundaries of areas to be cleaned rather than estimating subsurface averages. In some instances, though, estimating averages were necessary. For example, the criterion implies that the shallowest 5 em subsurface increment is 2.5 -7.5 em, but this interval was never sampled as such. Therefore, the average in this interval had to be estimated from 0-5 em and 5-10 em data. The method used to estimate the 2.5 - 7.5 em averageis deseribed in Tech Note 19.0. As the cleanup progressed, changes were made in the interpretation of various surface criteria. For more details concerning these changes, see Section 2.2.4. Both the area averaged over and the acceptable average value were altered. This meant that all the statistical analyses had to be flexible enough to allow estimates to be made for different sized areas and compared to various criteria levels. Fortunately, the kriging technique is quite flexible, so the original 50 m data could still be used. In those areas with 25 m data, it was relatively straightforward to compute the arithmetic means for various size areas. 5.3 DATA HANDLING Data handling responsibilities during the Enewetak cleanup project included not only statistical analyses but also data base management, data quality assurance and preservation, and the display of results in clear, useful forms. The types of information involved included not only raw data and final results, but also intermediate results, narrative descriptions of statistical methods, documentation for computer programs, ete. The onsite DRI statistician, assisted by the Navy data technician, had primary responsibility for data responsibility of DRI-Las Vegas. handling on-island. 148 Long-term data preservation was the