20 percent chanoe in the petroleum hydrocarbon concentration at that same 

 station in Phase II. The standardized range t = ( ymax - umin)/S=(180-120) 

 /53= 1.132, and the number of samples required with the a and g values set 

 at 10 percent is, from the a = 0.10 table, 14. (Quadratic interpolation 

 must be used with these tables.) If this number of samples is prohibitively 

 large, then the sensitivity requirements must be reduced and/or the error 

 risks must be increased. If the analyst accepts this number of samples, he 

 is estimating that there is a 90 percent probability that a 20 percent change 

 in the Phase I concentration of petroleum hydrocarbons at that sampling 

 station will be detected if 14 samples from that same station are analyzed 

 in Phase II. 



The number of samples generated from the extensive Phase I sampling pro- 

 gram proposed in the preceding sections could be very large. One means of 

 reducing the number of samples actually analyzed is to subject the samples 

 to preliminary analysis for the presence of the spilled substance. The 

 results of such preliminary screening could then be used to eliminate un- 

 promising sample sets. Samples that are not selected for analysis should 

 be stored for possible future analysis, not discarded. 



Another method of reducing the number of samples analyzed is sample 

 pooling. For instance, once an estimate of variation within a particular 

 sample type is obtained in Phase I, the remaining samples of that sample 

 type taken at similar sampling stations could be pooled prior to analysis. 

 Pooling of samples should, however, be done with discretion, especially if 

 the data generated from the pooled samples could possibly be used in court. 

 The reason for this is that when samples are pooled the effects of the 

 variation within the set of samples on the data are reduced at the expense 

 of losing the capability of measuring the variance within that set of 

 samples. This severely limits the range of possible statistical treatments 

 of the data thereby reducing both its scientific and legal values. Another 

 problem with sample pooling is that it usually requires that subsamples be 

 taken from a homogeneous mixture of the pooled samples. The fact that cer- 

 tain sample types, such as sediments (Chesler et al. 1976), do not readily 

 form homogeneous mixtures when pooled could introduce a significant addi- 

 tional source of error into the analysis. 



The same procedures described previously for estimating the number of 

 samples required to achieve a specified level of analytical precision can be 

 used to estimate the optimum number of samples for each sample pool. 



Since sampling in Phase II should be more selective than in Phase I, 

 fewer samples should be collected at this stage. Pooling of samples in 

 Phase II, then, should not be considered unless there are severe resource 

 limitations. 



Sampling Patterns 



The sampling pattern for each individual spill should be determined only 

 after the spill site has been viewed firsthand. As a selection guide, general 



184 



