550 



Fishery Bulletin 101 (3) 



greater power results in lesser sample size requirements, it 

 appears that species with higher trawl survey uncertainty 

 require fewer submersible survey samples. The reason 

 fewer samples are required is that the effect size-index 

 threshold has been increased and, generally, fewer samples 

 are needed to detect larger effects. The key to understand- 

 ing this relationship is that effect size is related to SD( D 

 ,), but power is a function of that effect size in relation to 

 the uncertainty in the data (s ). Essentially, the greater the 

 effect size in relation to the uncertainty in the data, the 

 greater the power. As SD{ D^)/s increases, the level of reso- 

 lution that can be detected by the trawl survey decreases. 

 Thus, our choice to set the effect size-threshold ( the level of 

 bias we need to be able to detect) equal to the uncertainty 

 of the trawl survey estimator (Appendix II) created a trade- 

 off between the level of resolution of the hypothesis test 

 and the power to detect that level. This criterion was an 

 arbitrary choice; a different relationship to describe this 

 tradeoff would yield different results. In practice, the rela- 

 tive level of acceptable bias versus precision will depend on 

 particular management objectives. 



To obtain sample size guidelines for estimating the trawl 

 survey habitat bias, we calculated c/^ using estimated values 

 for SD(Di), s , and a range of assumed values for A^M for 

 selected groundfish groups. We used information from our 

 submersible survey to characterize the variability of den- 

 sity estimates within trawlable and untrawlable habitats 

 is ), and information from past trawl surveys to character- 

 ize the variability in trawl survey estimates of abundance 

 (SD(0,)). The trawl survey .statistics used were derived from 

 the 1998 survey estimates available for the US- Vancouver 

 International North Pacific Fisheries Commission (INPFC) 



area shallow stratum (55-183 m) (Shaw et al., 2000), which 

 encompasses the study area location. By substituting the 

 calculated rf^ for d in Equation 1, we solved iteratively for 

 sample size (n ) using Excel Solver (Excel 2000 vers. 9.0.2720, 

 Microsoft Corp., Redmond, WA). The sample sizes obtained 

 provide guidelines so that a similarly designed study will 

 have an xVc chance (e.g. power of 0.80=80% chance) of de- 

 tecting a difference in mean density at least as large as the 

 random noise inherent in the trawl survey density estimator 



Results 



Submersible survey 



Sixteen dive sites were sampled — eight in each habitat type 

 (trawlable and untrawlable) (Table 1). In total, an estimated 

 85,900 m'-^ was covered across all sites. The untrawlable 

 sites (90-118 m) tended to be somewhat shallower than 

 the trawlable sites ( 106-148 m); however, we assumed that 

 this difference had little effect on fish density and species 

 composition within the study area. In general, we were not 

 successful in obtaining useful transect plots or reliable dis- 

 tance-traveled information with the WinFrog navigational 

 software package; however, we found the Trak-Point acous- 

 tic tracking system to be useful for obtaining the location of 

 the submersible with respect to the ship's position. We used 

 this information, together with the subsea communication 

 system, to guide the submersible along the predesignated 

 transect segments at each dive site. 



Our video survey largely confirmed our a priori assign- 

 ments of trawlable and untrawlable habitat (Table 2). At 



