Diamond: Estimation of shrimp trawl bycatch 



497 



tors in several significant 2-way and 3-way interactions 

 for the CPUE mean-per-unit estimator with delta lognor- 

 mally distributed catches of fish and shrimp. All four of 

 the ratio estimators were extremely sensitive to variance 

 in the auxiliary variable, which was either shrimp catch 

 or a variable measure of effort such as hours fished. Often 

 in field data, the variance in the catch of shrimp will be 

 much greater than the variance in the measure of effort; 

 therefore the basic F;S ratio method has both the greatest 

 bias and the highest variance in the auxiliary variable, 

 making it the least desirable of the five methods tested. 

 Both of the grand ratio methods were very sensitive to 

 observer coverage for normally distributed data, although 

 because of interactions with other parameters, it was hard 



to discern how increasing the number of observations 

 changed the bias of the estimates (Fig. 2). Surprisingly, 

 the correlation between the catch of shrimp and the catch 

 of fish in the normally distributed simulations was only 

 a significant main effect for the basic F:S ratio estima- 

 tor, and even for that estimator the CV of shrimp catch 

 had a much more profound effect. In the grand F:S ratio 

 estimator, there were two 3-way interactions between the 

 correlation coefficient and other parameters, but the other 

 variables seemed to exert much more influence on the by- 

 catch estimate than the correlation between the catches 

 offish and shrimp. 



Comparisons of bycatch estimators between the normally 

 distributed data and delta lognormally distributed data 



