552 



Fishery Bulletin 101(3) 



A summary of counts for large (>20 cm) and small (<20 cm) 

 fish is shown in Table 4. Small flatfish and rockfish were 

 very difficult to count, often becoming indistinguishable 

 from the background when the videotape was paused, 

 and their counts are most likely underestimated. Among 

 the large fish, "total rockfish" as a group was the most 

 abundant numerically followed by "total flatfish" as a 

 group. Of the large rockfish identified to species (Table 

 5), rosethorn rockfish were the most abundant followed 

 in order by yellowtail, greenstriped, yelloweye, tiger, and 

 redstripe rockfish. Unidentified rockfish represented 30% 

 of the total large rockfish enumerated. Of the large flatfish 

 identified to species (Table 6), Dover sole were most abun- 

 dant followed in order by arrowtooth flounder and Pacific 

 halibut. Unidentified flatfish represented 78% of the total 

 large flatfish counted. Other individual fish species and 

 groups identified below the generic classification level were 

 dominated by eelpout (Zoarcidae), raffish , skates and rays 

 (Raja), and greenling {Hexagramrnos spp.) (Table 7). 



Species composition differed considerably between habi- 

 tats. The number of individually identified species was 15 

 in the trawlable habitat, and 18 in the untrawlable habitat 

 (Table 8). Flatfish dominated in the trawlable habitat, and 

 rockfish in the untrawlable habitat. Yelloweye, redstripe, 

 silvergray, and quillback rockfish, as well as greenling 

 and wolf-eel were observed in the untrawlable habitat 

 but not in the trawlable habitat. Spiny dogfish (Squalus 

 acanthias), Pacific cod (Gadus macrocephalus), and salmon 

 (Oncorhynchus spp) were observed in the trawlable habitat 

 but not in the untrawlable habitat. 



Comparisons of fish densities and variances between 

 habitat types were made only for fish >20 cm in length 

 and in taxonomic units where reliable identification and 

 enumeration could be assured (Table 9). Thus, density com- 

 parisons were performed at the species level for distinctive 

 species (i.e. lingcod, yelloweye rockfish, and tiger rockfish), 

 but were made at the group level for "all rockfish" and "all 

 flatfish" bwcause of the presence of fish that could not be 

 identified to individual species within each of these groups. 

 For all comparisions, tests of homogeneity of variance of 

 fish density between habitats (//(,: s'^i=s^^) were rejected 

 using Cochran's test (Winer, 1971 ) ( «=0.05, k=2, df=7), indi- 

 cating heteroscedastisity (Table 9). Significant differences 

 in densities between habitats were found for each of the 

 species and group comparisons using the Mann-Whitney 

 two-sample test on ranks (Winer, 1971) (a=0.05, 2) (Table 

 9). Densities were higher in the untrawlable habitat for the 

 "all rockfish" group, tiger rockfish, yelloweye rockfish, and 

 lingcod; densities were higher in the trawlable habitat for 

 the "all flatfish" group. 



Statistical power analysis 



The validity of our approach for analyzing the statistical 

 sampling power of the submersible survey depends upon, 

 among other things, fidelity to the assumptions of the 

 two-sample t-test of means. The <-test requires that 1) the 

 two sample means are estimated from random samples 

 drawn from normally distributed populations, and that 

 2) the variance of the two populations are equal. Because 



our estimates of variance differed considerably between 

 habitats (Table 9), we examined the properties of our data 

 in more detail to confirm the reliability of using the t-test 

 for our statistical power analysis. We conducted a bootstrap 

 simulation experiment, in which we compared estimates 

 of empirical power derived from our study (n=8) with the 

 estimates of power obtained with Equation 1, under the 

 assumption of asymptotic normality. The results of this 

 comparison indicated that estimates of statistical power 

 obtained from Equation 1 were generally conservative 

 (indicated lower power) in relation to the empirical esti- 

 mates of power for simulated known differences in density 

 (Fig. 4). Given this result, we proceeded with our power 

 analysis based on the <-test, under the assumption that, 

 based on our observations, this approach will tend to err in 

 the conservative direction; that is, it will tend to understate 

 statistical power 



It is evident that, as it becomes necessary to detect small- 

 er effect sizes, the required sample size increases accord- 

 ingly. The relationship between sample size (;!=the number 

 of sample units Isubmersible dive sites] in each habitat 

 t3TJe) and the effect size-index (d) for density comparisons 



