506 



Fishery Bulletin 101 (3) 



The second phase of the experiment began with random 

 tows in an east to west direction. Complete location and 

 CPUE data for both species are located in Appendix I. In 

 order to analyze all alternatives, the lowest alternative 

 was used in the field for adaptive sampling during the 

 second phase, which resulted in the 220 kg/km criterion 

 value for POP from alternative 2. For SR-RE, the criterion 

 value was the mean CPUE of 418 kg/km from alternative 

 3. The remaining alternatives were simulated following the 

 completion of the survey. 



POP results 



After the initial tows, 25 random tows were selected for the 

 return trip across the area. All 25 were completed, of which 

 six became networks of more than one unit. A total of 106 

 tows were completed in the POP stratum. At one of the 

 tows that exceeded the criterion value, the captain deemed 

 that further adaptive sampling was not feasible because of 

 the presence of coral. Of the six networks, two overlapped, 

 resulting in five distinct networks. In these networks, 81 

 adaptive samples were taken, of which 49 exceeded the 

 criterion and 32 did not and were therefore edge units and 

 not included in the sample estimates. 



We compared the results of the original adaptive sample 

 ( alternative 2 ) with the simulated results of higher criterion 

 values (Table 2). The precision of simple random sample 

 estimates with both n (number of random samples) and v' 

 (number of random samples plus the number of adaptive 

 network samples, not edge units) was contra.sted with that 

 of the adaptive estimators described above. As the criterion 

 value increased, n remained the same, whereas v' and r (the 

 number of networks) decreased. At the 220 kg/km criterion 

 value (alt. 2), there were substantial reductions in SE over 

 the SRS estimators by using ACS estimators for both the ;; 

 and v' sample sizes. The 250 kg/km criterion value (alt. 3) 



resulted in a nearly identical sample to that of the 220 kg/ 

 km (alt. 2) criterion value and the loss of only one network 

 sample. Hence, the estimates were nearly identical. The HT 

 mean estimates were slightly lower than the HH estimates 

 for the two lowest criterion values (alts. 2 and 3) because 

 two networks overlapped. These networks became separate 

 at the next higher criterion value, which aligned the estima- 

 tors. The next highest criterion value of 540 kg/km (alt. 1) 

 showed that even though the sample size was reduced by 19 

 tows from the original criterion value, the ACS estimators 

 performed nearly as well, yielding just slightly larger SEs. 

 When the criterion was arbitrarily doubled to 1080 kg/km, 

 the sample size was further reduced by seven, and had 

 similar SEs to the 540 kg/km criterion value. 



The SRS and ACS bootstraps for POP resulted in very 

 different distributions. Five thousand replications showed 

 that the SRS distribution was bimodal and right skewed 

 (Fig. 3). The SRS mean fell on the second mode, which is 

 more than twice the ACS mean. This bimodal distribution 

 is driven by the presence of the very large random catch 

 (tow no. 60). If that haul is present in a bootstrap repli- 

 cate, then the SRS estimate tends to be high, leading to the 

 second mode in the bootstrap distribution. The ACS boot- 

 strap distribution was symmetric and closely resembled a 

 normal distribution (Fig. 3). The average estimates of bias 

 showed that the bias of HH was +4^( and the bias of HT 

 was -I7c. The standard error had an estimated bias of -t-S'/c 

 for HH and HT 



The results from this POP study and the previous 1998 

 study were both greatly affected by one or two very large 

 catches, as we expected for a highly clustered population. 

 Of interest is what happened when the largest catch was 

 changed to a nominal catch that still exceeded the criterion 

 value. Appendix II shows the results of changing haul no. 

 60 from 12,000 kg/km to 540 kg/km. In the comparison at 

 v', SRS outperforms ACS in terms of SE. However, it also 



