induced factors, but could just as easily be a result of this change in 

 technique. Thus, while the data collected between March 1973 and October 

 1977 are appropriate for detecting changes in that time period (perhaps 

 related to the operations of Units 1 and 2) , a more appropriate baseline 

 for the detection of quantitative changes related to the operations of 

 Unit 3 would begin with the October 1977 samples. Qualitative comparisons 

 that utilize the relative proportions of the species would still be 

 appropriate over the whole period. 



The frequency distributions of the catches of 12 selected trawl- 

 caught species were examined after first restricting the data to those 

 time periods and stations in which the species of interest was found 

 (Table 6). The distributions shown in Fig. 2-3 are exemplary. These 

 highly skewed, non- normal distributions typically had variances that 

 were related to the means. Eight out of 12 of the distributions approxi- 

 mated a negative binomial distribution (Poole 1974) (see Table 7) . Such 

 data can be normalized with the transformation: In (catch + k/2) . 

 Thus, all parametric analyses utilized the appropriately transformed 

 data from individual tows collected since October 1977. 



Limits of detection and sample size were evaluated and found to be 

 quite good for some species. The current program (consisting of three 

 replicates taken at six stations 26 times a year) provides 468 samples 

 per year. Considering winter flounder (Table 8), this sampling program 

 could detect a 20% (R=1.20) change in the means of the transformed data 

 (vs Ha: R=l) at a =0.05, 90% of the time ( 6 = . 90) . Detecting a 10% 

 change at these a and B levels would require 1195 samples per year. For 

 scup, 130 samples are currently collected from June through November at 

 NB, IN, and JC. At this level a 50% change could be detected ata=0.05, 



34 



