Diamond: Estimation of shrimp trawl bycatch 



491 



To statistically analyze the overall biases shown by 

 each estimator regardless of fishing conditions (i.e. mean 

 catches of fish or shrimp, CV, etc.), I first used paired 

 sample /-tests (SAS v. 8, The SAS Institute, Gary NC) to 

 separately compare each of the five estimates of fleet by- 

 catch with the "actual" bycatch for that fleet, based on the 

 following equation: 



where % bias 



estimated bycatch^ ^^^ 



"actual" bycatch , 



estimated bycatch, ,„ ^ - "actual" bycatch, ,, x 1 00 ( j^ 7 ) 

 "actual" bycatch,^ 



„ I, = the bias in the m"^ estimator for 



the 6'h fleet; 

 ,„ f, = the bycatch of the i^^ species by 



the w'*' estimator for the b^^ fleet; 



and 



the simulated actual bycatch of 



the ;"^ species by the 6"^ fleet. 



For these statistical tests, all fleets with normal distribu- 

 tions offish and shrimp were combined and analyzed sepa- 

 rately from the fleets with delta lognormal distributions of 

 fish and shrimp, giving sample sizes of 21,600 fleets for the 

 normal distribution and 118,810 fleets for the delta lognor- 

 mal distribution (Table 1). To look for significant factors 

 influencing the bycatch estimates for each of the five esti- 

 mation methods, I used ANOVAs on all main effects and all 

 2-way and 3-way interactions of the main effects for each 

 estimator. Although 7-way interactions were possible in the 



normally distributed simulations and 8-way interactions 

 were possible in the delta lognormal simulations (Table 1 ), 

 I stopped the analysis at 3-way interactions because of the 

 difficulty in interpreting higher level interactions. Main 

 effects were the following: mean catches offish and shrimp, 

 CVs of fish catches, CVs of shrimp catches, CVs of hours 

 fished, number of observations or observer coverage, cor- 

 relation coefficient in the normally distributed simula- 

 tions, and the probabilities of catching fish and shrimp 

 in the delta lognormal simulations. In these ANOVAs, the 

 response variable was the percent bias for each method, as 

 calculated above. 



Results 



Field sampling 



I observed 16 tows from five trips in Pamlico Sound 

 between July and October 1995 and 24 tows from five trips 

 in the Cape Fear River between August and October 1995 

 (Table 2). According to the 1995 trip tickets, these months 

 comprised the peak of the summer brown shrimp and fall 

 white-pink shrimp seasons; 77% of the total shrimp catch 

 and 75% of the total trips in the northern region and 63% 

 of the total shrimp catch and 54% of the total trips in the 

 southern region occurred during those months. All observed 

 tows were daytime tows, which is when fishing generally 

 occurs in these areas. All nets sampled in July were 2-seam 

 or 4-seam flat trawls, designed to catch brown shrimp, and 



