However, the log transformation was found to not normalize the data 

 (Table 6). Additionally, based on the results in Table 7, it was concluded 

 that the seine catch data did not come from a negative binomial distribu- 

 tion. 



Even though the data failed to meet the assumptions of normal 

 theory testing, these tests are fairly robust with respect to non- 

 normality and thus limits of detection and sample size were determined 

 using log (CPUE + 1) transformed data. The current program (consisting 

 of three replicates taken at six stations eight times a year) provides 

 144 samples per year. Considering silversides (Table 10) , this sampling 

 program could detect a difference of the means of the transformed data 

 of d=0.146 (40% change) (vs Ho: d=0) at ct= 0.10 70% of the time, if the 

 data were normally distributed. Detecting a d of 0.041 (10% change) at 

 a=0.10, 8=0.70 would require 322, 319, 233, 736, 1732 and 1556 samples 

 per year for sand lance, fourspine stickleback, Atlantic menhaden, 

 killifishes, silversides and total catch respectively. This difference 

 has considerably different meanings depending on the absolute size of 

 the catch. For example, a d=0.114 could represent the difference between 



CPUE=10 1,114 and CPUE=10 1,0 (13-10=3 fish) or between CPUE=10 • 11 and 



2 

 CPUE=10 * (130-100=30 fish) but still be a 30% change. Since station 



and time of year combined contribute more than half of the over all 



variance for all species considered, being able to detect a 50% change 



in the annual geometric means at a=0.10, 8 =0.80 or a =0.05, 8 =0.70 



may represent a respectable detectability. 



The present seining program costs approximately 360 personnel-hours 



per year to complete. Over the last 11 years approximately 17,590 fish 



have been caught per year and this amounts to about 0.02 hrs/fish. 



70 



