Ellis and DeMartini: Video camera sampling of Pristipomoides filamentosus abundance 



73 



(TOTTM) and time to first appearance (TFAP) of the 

 respective species (Table 2, LM form). The duration 

 of squid bait (BTM index) was significantly corre- 

 lated with the MAXNO index and the other video 

 indices for opakapaka but was more strongly corre- 

 lated with the MAXNO index for puffers (Table 2). 

 Videos indicated that puffers were usually respon- 

 sible for the removal of the squid bait; a direct rela- 

 tionship between puffer numbers and the rate of bait 

 disappearance was evident. Spearman's rank corre- 

 lations mirrored the parametric correlations. 



After log-transformation, the data pairs were ap- 

 proximately bivariate normal. Among all the video 

 indices, MAXNO was best correlated with InCPUE 

 (LLNO) for opakapaka (Table 2). The ML and LM 

 forms of the MAXNO video index were compared 

 separately with the longline CPUE, and the LM form 

 provided a slight but consistently better Pearson's 

 correlation than did the ML form for both opakapaka 

 and puffers. Therefore, the LM form of the MAXNO 

 index was used for all further parametric compari- 

 sons and analyses. 



The MAXNO-CPUE relationship was approxi- 

 mately linear (Fig. 4A), and its residual plot showed 



neither discernible pattern nor slope (P=1.0, Fig. 4B). 

 If all double-zero data are deleted, the correlation 

 between video MAXNO and longline CPUE loses sig- 

 nificance (r=0.55, P=0.08, n = ll). However, the 

 double-zero data were retained in subsequent analy- 

 ses because there was no a priori reason to believe 

 they did not represent real absences. 



The observed magnitude of hook loss ( x =32%) in- 

 dicates that longline CPUE is fundamentally inac- 

 curate and biased for sampling this habitat and spe- 

 cies assemblage. Apparently, most hook loss occurred 

 when puffers bit through the leader above the hook. 

 Hook competition is often a problem with longlines 

 when hooked fish begin to saturate available hooks 

 (Rothschild, 1967). Removal of hooks has a similar 

 effect. A multiple linear regression with two descrip- 

 tive variables, a puffer factor (Xj) equal to the num- 

 ber of hooks lost plus puffer catch and opakapaka 

 catch (X 2 ), was run to determine the effect of puffers 

 on the relation between longline CPUE and the video 

 MAXNO index for opakapaka. X : and X 2 were first 

 determined to be uncorrected (r 2 =0.02, P=0.62). The 

 model (Eqn. 5) for the multiple regression was forced 

 through the origin, because neither sampling device 

 could record the presence offish in its absence. The 

 total variation in the opakapaka video index ex- 

 plained by the model was 87% (i? 2 =0.87, P<0.001). 

 Opakapaka longline CPUE explained 83% of the varia- 

 tion (^=0.83, P<0. 001), and the puffer factor explained 

 an additional 4% of the variation (^=0.04, P=0.07). The 

 latter observation suggests that the puffer factor might 

 strongly influence video-longline relations for 

 opakapaka at times of relatively high puffer abundance. 



Precision for longline and video cameras was sepa- 

 rately examined. For both opakapaka and puffers, 

 cameras had nominally but consistently better pre- 

 cision (V=81% and 48%) than did longline CPUE 

 (V=91% and 71%). 



1 993 video statistics 



The MAXNO video index did not differ between shal- 

 low and deep positions (Student's r=0.27, P=0.79; 

 Kruskal-Wallis x 2 =0.09, P=0.76) in May 1993 (Fig. 5). 

 The mean MAXNO data lack a monotone trend over 

 stations (P=0.5), even though raw MAXNO values were 

 atypically large at several stations <x 2 =28.0, P=0.04; 

 Fig. 5). Overall, however, dependence among stations 

 was absent and neither precluded simple £-tests of the 

 means nor power estimates for tests of the means. 



Power analysis 



Power was estimated for opakapaka abundance in- 

 dices. Sample sizes for longline and video gear would 



